Introduction

Cerebral small-vessel disease (SVD) causes lacunar stroke, which accounts for a quarter of ischaemic stroke1 and is a major cause of vascular dementia.2 Despite its importance, the underlying pathogenesis is poorly understood. Radiologically, small lacunar infarcts are seen with or without more diffuse regions of white-matter lesions (WML), referred to as leucoaraiosis, and best seen as a high signal on T2- or fluid-attenuated inversion recovery (FLAIR) sequences.3 Two clinicopathological phenotypes of SVD have been hypothesised: multiple small lacunar infarcts with leucoaraiosis associated with diffuse small-vessel changes on pathology, and single or a few larger lacunar infarcts without leucoaraiosis associated with perforator vessel atherosclerosis.4 5

Considerable evidence suggests that, particularly for the leucoaraiosis phenotype, endothelial dysfunction plays a role in pathogenesis. Endothelial abnormalities have been shown on pathology.6 Circulating blood markers of endothelial activation are elevated, particularly in the leucoaraiosis variant.7 8 Such endothelial dysfunction could predispose to tissue damage via reduced cerebral blood flow and impaired autoregulatory responses; such abnormalities have been reported in SVD.9 10

In addition, endothelial dysfunction could cause blood–brain barrier (BBB) permeability increase.11 This has been hypothesised to result in leakage of plasma components into the vessel wall and surrounding brain tissue, contributing to vessel-wall damage and brain parenchymal damage, the latter resulting in leucoaraiosis.11 12 Consistent with BBB breakdown, immunohistochemical studies found extravasation of serum proteins.13 Furthermore, increased cerebrospinal fluid (CSF)/serum albumin ratio, a marker of BBB breakdown, was reported in vascular dementia.14 15 However, this technique does not allow regional variations in BBB breakdown, and differences between lesioned and normal-appearing white matter (NAWM), to be determined.

Recently, contrast-enhanced MRI has been used to study BBB permeability.16–19 An early study in vascular dementia suggested increased permeability, although another early study failed to replicate this finding.16 17 A recent study reported increased permeability in the white matter of patients with lacunar stroke compared with patients with cortical stroke.19 No studies have been performed using an automated analysis to determine permeability both within and outside WML. Within areas of leucoariosis any increase in permeability could be a secondary consequence of tissue damage. If increased permeability plays a causal role in pathogenesis, one might expect to detect increased permeability not only in areas of leucoaraiosis, but also in NAWM. Previous studies have not related permeability to the SVD phenotype and the presence of leucoaraiosis. If endothelial dysfunction is primarily important in patients with the leucoaraiosis subtype, one might expect BBB permeability to be related to the presence of leucoaraiosis.

We used MRI, with prolonged imaging following contrast injection, and determined MRI signal intensity changes that may relate to alterations in BBB permeability, in both NAWM and WML, in SVD and in normal controls. We examined the hypothesis that increased permeability would be related to the degree of leucoaraiosis, and would be present not only within WML but also in NAWM.

Methods

Subjects

Twenty-eight SVD patients were recruited. Inclusion criteria were a classical lacunar syndrome (pure motor stroke, pure sensory stroke, sensorimotor stroke or ataxic hemiparesis/clumsy hand dysarthria) with a compatible lacunar infarct on MRI. All patients were assessed by a consultant neurologist. Patients were included in the study at least 6 weeks after the last stroke or TIA to reduce the effect of acute ischaemia on permeability. Exclusion criteria were: age <18 years, stroke cause other than SVD (cardioembolic source, stenosis >50% in any extracranial or intracranial large artery, any cortical infarct, any subcortical infarct >1.5 cm diameter), and any contraindication to MRI. Twenty-one controls without any history of stroke, central nervous system disease or any other disease associated with white-matter damage were also recruited. These were spouses of patients and volunteers from a community-based study of normal ageing.

The study was approved by the local research ethics committee. All subjects gave written informed consent.

MRI scanning procedure

All MRI scanning was performed on a GE 1.5T Signa LX scanner equipped with 22 mT/m EchoSpeed gradients running 8x software (General Electric, Milwaukee, Wisconsin).

MRI included sagittal T1-weighted, axial FLAIR, high-resolution coronal 3D-fast spoiled gradient echo (3D-FSPGR), T1-weighted and axial 3D-FSPGR T1-weighted sequences. The high-resolution coronal 3D-T1-weighted imaging was used for image segmentation (TR/TE fixed at 11.1/5 ms, 1.1 mm slice thickness, no slice gap, field of view 28×28 cm, matrix 256×192, 176 slices). The imaging technique was similar to that used previously identify evidence of BBB permeability changes.19 For the dynamic contrast study, an axial 3D-FSPGR with flip angle at 12° was used (TR/TE 7.3/3.4 ms, 4 mm slice thickness, no slice gap, field of view 24×24 cm, matrix 256×256, 36 slices). The axial 3D-FSPGR T1-weighted imaging was used in a continuous multiphase mode, providing 36-slice images of the whole head at 27 time points over a 29 min 6 s acquisition time. A bolus of a gadolinium-based contrast agent (Omniscan, GE Healthcare Medical Diagnostics, Amersham, UK) was injected intravenously at a fixed dose of 40 ml per patient (5 ml/s) after the first image in this series. To correct for any scanner signal intensity drift, we used a linear mathematical correction derived from the MRI data of 11 healthy young subjects who underwent the 3D T1-weighted sequences without contrast application.

Image analysis

MRI data were analysed on an independent workstation (Sun Blade 100; Sun Microsystems, Mountain View, California).

Quantification of white-matter lesions and atrophy

WML load was determined using both a semiquantitative scale and a semiautomated programme. The semiquantitative Fazekas scale was used,20 with all scans being reviewed by the same experienced rater blinded to the permeability imaging results. An additional category was included to allow differentiation of more severe cases of leucoaraiosis. Leucoaraiosis was therefore rated as: 0=absent; 1=mild; 2=early confluent; 3=severe confluent.

For each patient, a ‘WML mask’ containing all WML was constructed by application of a semiautomatic region extraction technique using the contour function in the Dispunc image display program (David Plummer, University College London) to delineate the contours of areas of WML observed on FLAIR images.

To adjust for atrophy, brain tissue volume, normalised for subject head size, was estimated using SIENAX.21 22

Brain segmentation

Using the high-resolution T1-weighted images and the WML mask derived from FLAIR images, automatic brain segmentation was computed for each individual (figure 1). First, the T1-weighted images were segmented into grey-matter, white-matter and CSF tissue maps, incorporating a correction for image-intensity non-uniformity in SPM2.23 Second, the T1-weighted and FLAIR images were coregistered (using a 12-parameter affine transformation) to the precontrast 3D T1-weighted images in SPM2. These transformations were applied to the WML mask and the segmented images to coregister them to the precontrast 3D T1-weighted images. Third, hard segmentations of the coregistered segmented images were computed such that each image voxel was assigned to a single tissue class according to thresholds of intensity values defining the highest likelihood probability of each voxel belonging to the tissue types. The coregistered WML mask was then overlain on the hard segmentations to generate a brain mask image representing WML and NAWM at each image voxel.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Automatic segmentation technique. The precontrast T1-weighted image (a) was used as the target image for coregistration of the fluid-attenuated inversion recovery (b) and high resolution T1-weighted image (c). The final brain mask for the patient (d) is shaded according to tissue classification (dark grey, cerebrospinal fluid; mid-grey, grey matter; light grey, normal appearing white matter; white, lesion).

Analysis of dynamic contrast imaging data

For each of the 27 time-points after contrast injection, the 3D T1-weighted images were aligned to the same image prior to gadolinium injection using a 12-parameter affine transformation in FLIRT (FSL tools http://www.fmrib.ox.ac.uk/fsl/).24 The percentage signal change between the T1-weighted intensities pre- and postcontrast injection was computed at every voxel of each of the 27 time-points. Using the brain mask computed from the FLAIR and high-resolution T1-weighted images, the signal intensity change was determined in each of the segmented compartments of the brain. The mean percentage signal intensity change for each compartment was then plotted against time to calculate the area under curve values (AUC) over the full time course of dynamic imaging as a robust model-independent measure to detect differences in the vascular characteristics of each patient with high sensitivity.

Although signal intensity measurements over large volumes of a single tissue type increase sensitivity to detect small signal changes, tissue segmentation may not perfectly separate tissue signal from that of large blood vessels. To reduce partial volume effects, we also calculated the difference between average signal changes for control and patient data at each time point, on the assumption that the large vascular structures in the two groups would be similar; hence, partial volume effects would cancel. In addition, this difference measurement also allows cancellation of systematic errors from drift in scanner sensitivity that occurs over a long acquisition period.

Regions-of-interest analysis in the internal carotid arteries

Because slower washout of the contrast agent could reflect BBB damage but just as well slowed tracer clearance due to, for example, impaired renal function, we assessed AUC in the internal carotid arteries of each individual by manually placing small regions of interest over the internal carotid arteries, as suggested and performed by Wardlaw et al.19 The regions of interest arterial blood data were included in the multivariate analyses (outlined below) to adjust for arterial tracer concentration.

Statistical analysis

For univariate analyses of categorical variables, we used a two-tailed Fisher exact test or Pearson χ² test as appropriate. For continuous variables, differences between groups were assessed by the two-tailed independent samples t test (data with normal distribution), Mann–Whitney U test (skewed data, two groups) and Kruskal–Wallis test (skewed data, more than two groups). Correlation of continuous variables was tested using the Spearman r. To identify independent predictors of changes in the AUC (ie, the postcontrast signal characteristic) in NAWM and cerebrospinal fluid, multivariate regression analyses were performed by entering the following variables into the model: age, gender, weight, brain volume, leucoaraiosis grade, AUC in the internal carotid arteries (ICA), presence of hypertension, presence of diabetes and current smoking. p Values <0.05 were considered to be statistically significant. SPSS software (Version 14.0; SPSS, Chicago) was used for all analyses.

Results

Baseline characteristics

Baseline characteristics are given in table 1. Four patients were excluded prior to image analysis: one because postcontrast scanning had to be stopped immediately after injection of gadolinium due to severe nausea; two due to movement artefacts and poor image quality; one due to incomplete postcontrast T1 acquisition due to technical reasons. This left 24 patients for analysis. Three of the 21 controls were excluded, due to movement artefacts and poor image quality in two, and incomplete postcontrast T1 acquisition for technical reasons in one. Three further controls were excluded because they had confluent leucoaraiosis (grade 2). Therefore, 15 controls were included in the analysis

Demographic characteristics are shown in table 1. There was no difference in mean (SD) age between cases and controls (67.1(±8.9) and 65.5(±6.7) respectively) or in gender or weight. Hypertension and hypercholesterolaemia were more common in cases (table 1). In patients, the median time delay between MRI and the last TIA or stroke was 12 months (range 1.5–56).

Kinetics of signal intensity change

The average kinetics of signal intensity change in white matter in both cases and controls are shown in figure 2 and are similar to that expected from changes in plasma contrast agent concentration.25 The difference plots in figure 3 demonstrate differences between control and patient long time constant kinetics for NAWM, but not for WML. The increase in the signal intensity difference between patients and controls for NAWM is suggestive of the greater accumulation of contrast in the patient group compared with controls, in whom the contrast agent is not expected to cross the BBB, and therefore suggests increased leakage.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Kinetics of postcontrast signal intensity change over time in patients and controls in the cerebrospinal fluid (graph a), normal appearing white matter (graph b), and white-matter lesions (graph c). Standard error bars are marked on the graphs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Kinetics of the postcontrast signal changes shown as the difference between the signal averages of the control and patient group data shown in figure 3 for cerebrospinal fluid (graph a), normal appearing white matter (graph b) and white-matter lesions (graph c). Error bars represent the square root of the sum of squares of the standard errors on the patient and control groups.

Patients had significantly higher AUC values for NAWM compared with controls (Mann–Whitney U, median; range: 1.0; 0.23–1.65 vs 0.76; 0.3–1.36; p=0.033) but there was no difference for WML (p=0.966), ICA (p=0.172) and CSF (p=0.258). In patients, there was no correlation of the time from the last stroke or TIA event and the signal intensity change for NAWM (Spearman r=−0.064, p=0.82).

Analysis according to leucoaraiosis grade

To study the relationship between AUC and degree of leucoaraiosis, subjects were assigned to the following four groups: controls with LA grade 0–1 (n=15); patients with LA grade 0–1 (n=8); patients with LA grade 2 (n=6); and patients with LA grade 3 (n=10). Kruskal–Wallis tests comparing AUCs between groups (figure 4) showed that a higher grade of LA was associated with increased AUC in NAWM (p=0.039), but there was no trend across subgroups for WML (p=0.641).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

Plots of area under curves (AUC) of postcontrast signal intensity change in subgroups according to leucoaraiosis grade and subject status in cerebrospinal fluid (graph a) normal-appearing white matter (graph b) and white-matter lesions (graph c). A higher AUC value is consistent with a higher BBB permeability. The mean values are shown to the right of subgroup data with standard error bars marked. p Values represent Kruskal–Wallis test results comparing AUC values between subgroups.

Determinants of contrast agent leakage

Multiple regression analyses were performed to assess the relationship between clinical variables and AUC in NAWM and CSF in the whole dataset (table 2). After controlling for age, gender, weight, brain volume, AUC in the internal carotid arteries, hypertension, diabetes and current smoking, leucoaraiosis grade was found to be an independent predictor of leakage as represented by the AUC in NAWM (p=0.027) and in CSF (p=0.001) (table 2).

Discussion

We have applied a sensitive MRI technique to investigate small differences in global perfusion and permeability characteristics of brain tissue. Furthermore, we used an image-segmentation technique to generate tissue masks from which postcontrast AUCs were computed of signal intensity change over an extended time period. Comparison of AUC provides an estimate of blood–brain barrier permeability alterations.

Our results are consistent with increased BBB permeability in the white matter of patients with SVD. The AUC in NAWM was greatest in those SVD patients with leucoaraiosis, and there was a positive correlation between AUC and degree of leucoaraiosis. This supports the hypothesis that BBB permeability is particularly important in the leucoaraiosis subtype of SVD. These correlations persisted after controlling for conventional cardiovascular risk factors, and on multivariate analysis leucoaraiosis grade remained a significant determinant of AUC. A relationship was also found between leucoaraiosis grade and CSF AUC, consistent with increased contrast leakage into the CSF in the presence of more severe leucoaraiosis.

We used an automated segmentation approach to separately determine AUC in established WML and in NAWM. We found a progressive increase in AUC in the NAWM of the patient group with leucoaraiosis grade. This would be consistent with increased permeability preceding the development of WML lesions and thus playing a causal role. Once developed, WML in both symptomatic and asymptomatic subjects had similar permeability characteristics as described by the AUC.

In the signal intensity with time plots (figure 2), there is initially a step increase, followed by a slow decay. Such a pattern is characteristic of a large postcontrast signal intensity increase due to the vascular compartment of the signal, followed by a decline due to a combination of contrast wash-out and contrast leakage into the tissue compartment of the signal. For NAWM, there are significant differences between patients and controls in the difference plots (figure 3) and the AUC (figure 4) due to changes with a long time constant. As no differences between the two groups were observed in brain volume (table 1) or in initial signal intensity changes (figure 3), it is unlikely that group differences in tissue/vascular partial volume are responsible for the observed group differences in AUC. Consequently, the greater AUC for NAWM in patients compared with controls is more consistent with an accumulation of contrast agent due to increased leakage out of the vascular space. However, although the AUC is a sensitive and model-independent measurement, its interpretation is not without ambiguity.26 In particular, simulations show that the AUC will increase with increased permeability, increased extracellular extravascular space or increased blood flow.27 Increased blood flow is unlikely to be an explanation for our findings. Patients with the highest AUC were those with the most severe leucoaraiosis, and cerebral blood flow is reduced in the NAWM in this group.9 Therefore, increased blood flow is an unlikely candidate for explaining the signal differences between patients and controls in NAWM. Furthermore, in a previous study with a similar acquisition protocol, signal enhancement declined more rapidly in the blood compared with the brain, and CSF signal enhancement (uncontaminated by intravascular enhancement) was quite different to that of the blood.19 Previous studies using diffusion-weighted imaging have shown an increase in apparent diffusion coefficients in normal white matter of patients with leucoaraiosis,28 which could be interpreted as an increase in the extracellular space. However, the postmortem characteristics of leucoaraiosis are of axonal loss accompanied by gliosis, factors which provide an explanation for the increased median diffusivity previously observed,28 but with no change in extracellular space.29

An alternative potential explanation could be that slower washout of contrast merely reflects slowed tracer clearance—for example due to impaired renal function in patients with leucoaraiosis. To control for this, we measured signal intensity in the arterial compartment by placing regions of interest in the carotid artery. The associations persisted after controlling for this, suggesting that slower tracer clearance does not account for the differences we observed.

We used a fixed dose of contrast rather than a weight-adjusted dose. Differences between cases and controls, and the relationships between AUC and leucoaraiosis grade persisted after including weight in the multivariate regression. However, future studies should include weight-adjusted contrast doses.

Our findings are consistent with results obtained in previous studies using biochemical techniques, primarily those measuring CSF/serum albumin ratios in order to identify increased BBB permeability. A recent meta-analysis concluded that, while studies were disparate and individually gave conflicting results, the overall data suggested that BBB permeability is increased in vascular dementia, for which SVD is a major cause.30 Data were more limited from patients with MRI-identified WML but were consistent with increased BBB permeability. However, the CSF/serum albumin technique does not allow regional differences in permeability to be determined, and differentiation of whether the increased permeability is a secondary phenomenon to tissue damage in WML, or an early phenomenon occurring in NAWM.

The few previous studies using MRI have given conflicting results. A small study on 10 demented patients with WML failed to detect any differences in contrast enhancement of WML and NAWM between patients with and without elevated CSF/serum albumin ratio.16 However, the sample size was not powered to detect moderate differences. In 17 subjects with Binswanger disease, permeability within WML was greater than in non-demented subjects (10 with cerebrovascular disease and 14 controls).17 Recently, using a similar acquisition technique to that used in our study, BBB permeability was significantly higher in the CSF of patients with lacunar compared with cortical stroke.19 There was a non-significant increase in white-matter permeability. No differences were found in the grey matter. However, the cortical stroke control group obviously had cortical pathology, and this may have obscured any increased GM permeability in the lacunar stroke group.

Previous MRI studies, with one exception,17 have not determined permeability separately in WML and NAWM. They have also not determined whether there may be differences in permeability between the different subtypes of SVD, and whether increased permeability is primarily seen in the leucoaraiosis phenotype. These data not only show clear differences in leakage characteristics between lesioned and non-lesioned white matter, but also suggest increased permeability in NAWM, supporting the hypothesis that (widespread) increased BBB permeability may be a potential pathomechanism by which tissue damage can occur. A primary disease causal event could be alteration of cerebral microvascular endothelial (ie, BBB) function. This could result in extravasation of plasma components into the arteriolar wall (causing the wall thickening, cellular infiltration, disintegration and perivascular damage seen pathologically) and the leakage into the adjacent brain parenchyma causing ‘perivascular oedema-related lesions’ which are then seen radiologically as leucoaraiosis.11

In conclusion, this study provides evidence consistent with increased BBB permeability in NAWM of SVD subjects with leucoaraiosis. These findings are consistent with increased BBB permeability playing a causal role in disease pathogenesis. Further data are required from larger studies to confirm these findings. However, only longitudinal studies using consistent quantitative contrast-enhanced MRI approaches in large enough numbers of subjects can determine whether BBB permeability in NAWM predicts the future development of WML in areas of abnormal leakage. If it indeed plays a causal role in disease, this may offer new treatment approaches.