Disruption of the Blood–Brain Barrier Predicts Progression of Cerebral Small Vessel Disease White Matter Hyperintensities

Abstract

Objective

The objective of this study was to test if blood–brain barrier (BBB) disruption, detected using dynamic susceptibility contrast (DSC) imaging, would predict progression of white matter hyperintensities (WMHs) over the subsequent year in patients with chronic cerebrovascular disease.

Methods

The study included patients with a history of stroke and at least early confluence of WMH. Magnetic resonance imaging (MRI) scans performed at baseline (> 3 months from stroke) and again 1 year later were segmented to calculate the WMH volume expressed as a fraction of the total brain volume. Change in WMH volume between the 2 timepoints and progression of WMH were the outcome measures. BBB disruption was measured using DSC imaging on the baseline MRI. WMH masks were dilated by 3 mm to create a mask of the adjacent normal appearing white matter (penumbra). BBB disruption was averaged within the WMH and the penumbra.

Results

Fifty patients were included; median age was 69 years, and 46% were women. The mean WMH fraction was 1.25% at baseline and 1.36% at 1 year. The mean baseline BBB disruption was 0.20% in the WMH and 0.22% in the penumbra. More severe BBB disruption was associated with greater WMH progression when measured in the WMH (ß = 0.95, confidence interval [CI] = 0.39–1.51, r ² = 0.19, p = 0.001) and in the penumbra (ß = 0.81, CI = 0.10–1.53, r ² = 0.10, p = 0.027). The best predictor of progression was BBB disruption in the penumbra with an odds ratio (OR) of 2 (OR = 2, CI = 1.01–3.96, p = 0.046) for each 0.1% increase in BBB disruption.

Interpretation

More severe BBB disruption was predictive of greater WMH progression in patients with chronic cerebrovascular disease. ANN NEUROL 2026;99:437–448

Cerebral small vessel disease (CSVD) is thought to represent end organ damage due to longstanding poorly controlled vascular risk factors, such as hypertension. ¹ Long before vascular cognitive impairment and dementia (VCID) develops as a clinical manifestation of CSVD, radiographic evidence of CSVD can be seen on magnetic resonance imaging (MRI) including white matter hyperintensities (WMHs), microbleeds, enlarged perivascular spaces, microinfarcts, and atrophy. ² The extent of WMH on T2‐weighted MRI is a biomarker of VCID risk, ³ and the expansion of WMH is a biomarker for cognitive decline ⁴ with patients who have a history of clinical stroke being at particularly high risk. ⁵ In the presence of WMH, control of vascular risk factors, such as hypertension, appears to slow, but not halt, the progression of WMH, ⁶ and has not been shown to prevent dementia. ⁷ Although risk factor management and secondary stroke prevention medications are used clinically to avoid exacerbating CSVD progression, a truly disease‐modifying treatment remains elusive.

The pathogenesis of CSVD is thought to stem from arteriolosclerosis and loss of vasoreactivity of the microvasculature, which disrupts autoregulation and endothelial cell function and results in intermittent hypoxia. ⁸ This is thought to induce a sterile, non‐immune cell‐mediated, inflammatory response that leads to a state of chronic progressive tissue injury. ⁹ , ¹⁰ , ¹¹ Early in this process, disruption of the blood–brain barrier (BBB) can be seen, ⁹ implicating it in the conversion of normal appearing white matter (NAWM) to WMH. In humans, BBB disruption, detected as an increased concentration of albumin in cerebrospinal fluid (CSF), is associated with cognitive decline. ¹² Thus, BBB disruption measured in the NAWM is a potential biomarker of disease activity and progression of CSVD.

Dynamic susceptibility contrast (DSC) imaging is a type of MRI technique that is commonly used to measure blood flow in the brain. However, DSC MRI is also capable of detecting BBB disruption. ¹³ DSC MRI has been used extensively to study BBB disruption in acute ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ and subacute ¹⁸ , ¹⁹ , ²⁰ stroke, but also appears to be able to detect BBB disruption associated with CSVD. ²¹ , ²² , ²³

To test whether BBB disruption detected with DSC MRI in patients with a remote history of stroke (> 3 months prior) would predict subsequent progression of WMH over 1 year, a prospective observational study was conducted (ClinicalTrials.gov ID: NCT03366129). The results for the primary outcome of that study are presented here, which tested the prespecified hypothesis that disruption of the BBB measured in regions of vulnerable NAWM would predict progression of WMH over 1 year.

Methods

Human Research Protocol

The Deputy Ethics Counselors and Institutional Review Boards of National Institute of Neurological Disorders and Stroke/National Institutes of Health (NINDS/NIH; IRB Protocol Number: 18 N0020), Suburban Hospital, Johns Hopkins Medicine, Bethesda, MD (Protocol Number: IRB 00156306), and MedStar Washington Hospital Center, Washington, DC (Protocol Number: 2018‐093) approved this study, and it is registered with clinicaltrials.gov (NCT03366129). All participants provided written informed consent.

Data Integrity and Availability

The corresponding author oversaw the integrity of data collection and analysis. Results are presented according to STROBE criteria for observational studies. Upon reasonable request to the corresponding author, the data will be made available after close‐out and publication of the parent study under a formal data‐sharing agreement and with approval from the requesting researcher's local ethics committee. The software used in the study is freely available for non‐commercial use (Freesurfer, FSL) or commercially available (Matlab, Stata).

Study Population

The study screened potential participants from 2 local hospitals but also allowed for outside referrals. The aim of the study was to recruit a population of patients at risk for post‐stroke cognitive decline who had a burden of CSVD that could be tracked over time. Participants were eligible for this longitudinal cohort study if they had a history of an ischemic stroke or transient ischemic attack (TIA) and demonstrated CSVD on a prior MRI scan in the form of at least early confluent WMH (Fazekas grade 2). ²⁴ Participants were excluded if they were unable to ambulate independently, could not communicate verbally in English, had concurrent neurologic disease predisposing them to WMH of non‐cerebrovascular origin, or had severe dementia based on a cognitive screening (Montreal Cognitive Assessment [MoCA] score of less than 14, or a 6‐item screener ²⁵ score of less than 4). Modified Rankin score was collected but not used as an inclusion/exclusion criterion. Clinical and demographic data were collected at the time of enrollment based on patient interview and chart review. Prior stroke events were classified based on Trial of ORG 10172 in Acute Stroke Treatment (TOAST) criteria. ²⁶ Enrollment could occur any time after the prior stroke or TIA, but enrolled participants had their first research MRI scan performed at least 3 months after their prior ischemic event. Although the study collected MRI scans at multiple timepoints, to complete the primary outcome of the study, participants had to have at least one additional MRI scan 1 year (± 30 days) after their baseline scan. All MRIs for each individual patient were performed on the same scanner. The target sample size for the primary outcome of the study was 50 patients.

MRI Protocol

MRI scans were performed at 2 hospital sites using either a 3T Siemens Skyra or a 3T Phillips Achieva. The acquisition included the following sequences: a 3D T1‐weighted (resolution 1 × 1 × 1 mm), a 3D fluid‐attenuated inversion recovery (FLAIR; resolution 1 × 1 × 1 mm), an axial FLAIR (resolution approximately 1 × 1 × 3.5 mm), diffusion tensor imaging (DTI; Siemens: 12 directional images with B = 1,000 s/mm², 4 B = 0 images, and resolution 1.7 × 1.7 × 3.5 mm interpolated to approximately 1 × 1 mm or Phillips: 15 directions at B = 1000 s/mm², 1 B = 0 image, and 2 × 2 × 3.5 mm interpolated to approximately 1 × 1 mm), and DSC perfusion weighted imaging (PWI; TR = 1–1.5 s, TE = 25–45 msec, 2 × 2 mm interpolated to in‐plane resolution of approximately 1 × 1 mm, 7 mm slice thickness, 20 slices, 40–80 dynamics), which was collected during a single injection of a weight‐based dose of gadolinium (0.1 mmol/kg) using either Gadavist (gadobutrol; Bayer Schering Pharma, Whippany, NJ) or Multihance (gadobenate dimeglumine; Bracco Diagnostics, Monroe Township, NJ) at a flow rate of 5 ml/s.

MRI Processing

For each patient and timepoint, 3D FLAIR and T1 were used to segment brain tissue into regions‐of‐interest (ROI) for gray matter, white matter, and CSF using Freesurfer's Samseg. ²⁷ FLAIR WMH were segmented from 3D FLAIR and T1 using a deep‐learning algorithm trained using DeepMedic ²⁸ on expert‐labeled MRIs from 47 research participants with mild to severe chronic cerebrovascular disease and then tested on a separate set of 18 participants from this study demonstrating moderate to high overlap compared to expert‐labeled semi‐automated manual lesion tracing (mean Dice similarity index = 0.68, SD = 0.11). Tissue and WMH segmentation masks were visually checked and manually corrected if needed. FLAIR hyperintensity due to chronic infarct could not be reliably distinguished from background WMH and so was included in the WMH ROI. FSL ²⁹ was used to align WMH and white matter ROIs with DSC images using DTI as an intermediate to minimize co‐registration errors between different pulse sequences and resolutions. DTI B0 was aligned to T1 using 12 DOF linear registration and then refined using nonlinear alignment of fractional anisotropy (FA) to T1. The reverse warp was calculated and applied to transform white matter and WMH ROIs into DTI space. DSC was motion corrected by rigid aligning each volume to the fourth DSC volume, and then subsequently to the DTI B0 image. The 2‐step transformations were combined and applied while only interpolating once.

To generate BBB images, the DSC 4D time series was processed to detect leakage of gadolinium through the BBB. Clinically, DSC imaging is used for bolus tracking during an injection of gadolinium contrast. The signal change over time is largely due to intravascular contrast moving through the brain (due to T2* effects) and is typically processed to create maps of cerebral perfusion. However, in the setting of contrast leakage through the BBB into the brain parenchyma, there is an opposing signal coming from the extravascular accumulation of contrast (due to T1 effects). ¹³ Thus, in the setting of BBB disruption, the recorded signal is affected in a manner that is proportional to the concentration of leaked contrast. ³⁰ After adjusting for regional difference in blood flow using an arrival time correction (ATC), ³¹ the fraction (or percentage) of signal change due to contrast leakage, referred to as K2, can be isolated resulting in a voxel‐by‐voxel map of BBB disruption. Figure 1 demonstrates how the recorded signal of a DSC acquisition is affected by contrast leakage through a disrupted BBB. All BBB values reported in this paper reflect this percent change in signal due to contrast leakage.

Figure 1.

BBB permeability imaging. This figure demonstrates how disruption of the BBB affects the recorded signal on a DSC MRI acquisition. The top left shows the BBB heatmap for the white matter of a patient in the study. The bottom left shows the color code indicating the severity of BBB disruption (K2) shown on the MRI. On the right, the average recorded signal is shown for each echelon of BBB disruption using a similar color code as the MRI. The dark blue line shows how the signal behaves in the absence of BBB disruption. Of note, the recorded signal has been converted to delta R (sometimes referred to as the gadolinium concentration curve) to reflect the change in signal from baseline. Additionally, an ATC has been applied to the curves prior to averaging them to remove the effect of regional differences in blood flow. ATC = arrival time correction; BBB = blood–brain barrier; DSC = dynamic susceptibility contrast; MRI = magnetic resonance imaging. [Color figure can be viewed at www.annalsofneurology.org]

The WMH penumbra was defined as the NAWM surrounding the WMH and was isolated by dilating the WMH lesions in‐plane by 3 mm (3 voxels at a resolution of 1 × 1 × 1) within the white matter mask and then subtracting the non‐dilated lesions (Figure 2). A 3 mm penumbra was used because new WMH progression is most likely to occur in this region. ³² Mean BBB disruption was calculated for each of the resulting ROIs (WMH and penumbra) by averaging the K2 values for all voxels within the ROIs. Figure 2 shows an example of the segmentations and BBB heatmaps for a patient that showed an increase in WMH volume over 1 year.

Figure 2.

WMH progression. The primary end point of the study was the change in WMH volume as a fraction of brain volume. A single slice of the brain is shown for 1 patient at 2 timepoints. The images in the blue box are from the baseline timepoint: the top right is the baseline FLAIR image demonstrating WMH; the top middle shows the segmentation of the WMH and top left shows the segmentation of the WMH penumbra, which is a 3‐mm dilation of the WMH segmentation. Bottom right shows the BBB heatmap overlain on the FLAIR scan; the bottom middle shows the BBB disruption just in the WMH, and the bottom right shows the BBB disruption just in the penumbra. The images in the pink box are from the 1‐year timepoint: the top left shows the follow‐up FLAIR scan, the top right shows the segmentation of the WMH for the timepoint, and the bottom left shows the new WMH detected over the 1‐year period. BBB = blood–brain barrier; FLAIR = fluid‐attenuated inversion recovery; WMH = white matter hyperintensity. [Color figure can be viewed at www.annalsofneurology.org]

All WMH volumes reported in this paper were normalized to be a percentage of the total brain volume measured at the same timepoint (WMH volume / [white matter volume + gray matter volume]). Total brain volume was calculated by adding the white matter and gray matter segmentations. To avoid the effects of encephalomalacia from the prior stroke (which may continue beyond 3 months), the analysis was restricted to the contralateral hemisphere (for both the numerator and the denominator) when the prior stroke was unilateral and supratentorial. Both hemispheres were used if the stroke was infratentorial, bilateral, or absent. BBB measurements were calculated by one author (R.L.) who was blinded to the results of the WMH volume analysis; WMH volumes were calculated by one author (K.C.K.) who was blinded to the results of the BBB analysis.

Statistical Analysis

Patient characteristics were summarized using descriptive statistics. The general linear model (GLM) approach was used to examine the association between the dependent variable of change in WMH over 1 year and the primary independent variables of interest, namely BBB disruption measured in the WMH and the penumbra. Baseline WMH volume was also examined as an independent variable given its known association with risk of WMH progression. ³ Covariates, including demographic variables of age and sex, and vascular/stroke risk factors, such as hypertension, hyperlipidemia, diabetes, atrial fibrillation, coronary artery disease, tobacco use, obstructive sleep apnea, as well as clinical factors including National Institutes of Health Stroke Scale (NIHSS) at the time of the stroke, time from the stroke to the first MRI, time between MRIs, interval ischemic event during the study, and the scanner site, were examined in unadjusted, univariable analysis, and accounted for using multivariable GLM based on the following prespecified criterion. A p value of < 0.1 in unadjusted analysis was used as the criterion for inclusion of a covariate in the multivariable analysis. Potential violation of GLM assumptions were examined through a series of residual‐based diagnostic plots.

We also used a logistic regression modeling approach to evaluate the odds of WMH progression associated with BBB disruption measured in the WMH and the penumbra and calculated the corresponding area under the receiver operator‐characteristic (ROC) curve (AUC) for these predictors of interest. The binary outcome of WMH progression was defined as any increase in the WMH volume at 1 year; this definition was chosen to ensure that even small changes in WMH volume would be included, given the relatively short duration of the study. However, this approach is susceptible to errors in volumetric calculations which could introduce noise. So, in a sensitivity analysis, we also tested an alternative definition of WMH progression using a higher threshold of 0.15 based on the upper tertile of the distribution of WMH volume change. Again, we used logistic regression to test whether BBB disruption in the WMH or penumbra was associated with WMH progression at 1 year and calculated the AUC for each BBB predictor. Analyses were carried out using Stata 17 and SAS 9.4, and a 2‐sided p value of < 0.05 was considered statistically significant.

Results

Clinical and Radiographic Characteristics of the Population

Of the 81 patients enrolled in the study, 50 completed the necessary imaging for the primary outcome and were included in this analysis. The median age was 69 years, and 46% were women. A prior (> 3 months from the study's baseline) qualifying ischemic event (referred to here as the index event) was stroke in 46 patients, TIA in 2 patients, and retinal artery occlusion in 2 patients. The median time between the ischemic event and first research MRI scan at the baseline timepoint was 4 months. The median number of days between the MRI scans at baseline and 1‐year follow‐up was 371 days (interquartile range [IQR] = 362–383 days). In patients with stroke, it was supratentorial and unilateral in 34 patients, infratentorial in 8 patients, and bilateral in 4 patients. The etiology of the index event by TOAST criteria ²⁶ was small vessel disease in 15 participants, cardioembolic in 12, large artery atherosclerosis in 11, cryptogenic in 10, periprocedural in 1, and vertebral artery dissection in 1 patient. The median NIHSS at the time of acute stroke was 2 (IQR = 0–4). The median pre‐stroke modified Rankin Scale (mRS) was 0 (IQR = 0–0), and the median mRS at the time of the baseline research MRI was 1 (IQR = 0–1). There were 11 patients who suffered an interval ischemic event during the follow‐up; 2 patients suffered a clinical stroke, 7 patients had an incidental acute or subacute infarct seen on diffusion MRI, and 2 patients had both a clinical stroke and an incidental infarct.

The mean percent of the brain that was part of the WMH was 1.25% (SD = 1.09) at baseline and 1.36% (SD = 1.24) at 1 year. The median change in WMH volume was 0.06% (IQR = 0.03–0.20). Increased WMH volume was detected in 34 patients (68%), and the mean change for those who progressed was 0.23% (SD = 0.33). The mean baseline BBB disruption was 0.20% (SD = 0.15) in the WMH and 0.22% (SD = 0.13) in the penumbra. Table 1 summarizes the participant characteristics.

Table 1.

Participant Characteristics (n = 50)

Age, yr: median (IQR)	69 (63–76)	Index Event TOAST Criteria
Female: n	23 (46%)	Small vessel occlusion	15 (30%)
Risk factors: n		Large artery atherosclerosis	11 (22%)
Hypertension	43 (86%)	Cardioembolic	12 (24%)
Hyperlipidemia	40 (80%)	Cryptogenic	10 (20%)
Diabetes	17 (34%)	Other known etiology	2 (4%)
Atrial fibrillation	15 (30%)	Index event severity
Coronary artery disease	8 (16%)	Median NIHSS (IQR)	2 (0–4)
Tobacco use	5 (10%)	Median pre‐stroke mRS (IQR)	0 (0–0)
Obstructive sleep apnea	17 (34%)	Median post‐stroke mRS (IQR)	1 (0–1)
Index stroke event		Interval ischemic event: n
Stroke	46 (92%)	Incident infarct	9 (18%)
Transient ischemic attack	2 (4%)	Clinical stroke	4 (8%)
Central retinal artery occlusion	2 (4%)	Imaging findings
Time since index stroke event		Mean WMH percent volume (SD)
Median days (IQR)	122 (105–233)	Baseline	1.25% (1.09)
Index stroke anatomic location		One year	1.36% (1.24)
Cortical	17 (34%)	Median change WMH percent volume (IQR)	0.06% (−0.03 to 0.21)
Subcortical/brainstem	24 (48%)	Mean BBB disruption (SD)
Cerebellar	4 (8%)	WMH	0.20% (0.15)
Cortical and cerebellar	1 (2%)	Penumbra	0.22% (0.13)
Eye	2 (4%)	Imaging site: n
Index stroke hemispheric location		Suburban Hospital	42 (84%)
Unilateral	34 (68%)	MedStar Washington Hospital Center	8 (16%)
Bilateral	4 (8%)
Infratentorial	8 (16%)

BBB = blood–brain barrier; IQR = interquartile range; mRS = modified Rankin Scale; NIHSS = National Institutes of Health Stroke Scale; SD = standard deviation; TOAST = Trial of ORG 10172 in Acute Stroke Treatment ¹ ; WMH = white matter hyperintensity.

Unadjusted Analysis: Predictors of Change in WMH Volume

Using unadjusted GLM analysis, larger amounts of WMH accumulation over the subsequent year was associated with more severe BBB disruption measured in the WMH at baseline (ß = 0.95, CI = 0.39–1.51, r ² = 0.19, p = 0.001), indicating almost a one‐to‐one increase in the percentage of WMH volume for each percent increase in BBB disruption. This relationship is shown in Figure 3A.

Figure 3.

Baseline BBB disruption predicts WMH progression. Scatter plots show the relationship between BBB disruption at baseline (x‐axis) and change in WMH volume fraction over the subsequent year (y‐axis). Plot (A) on the left shows this relationship for BBB disruption measured within the WMH, and plot (B) on the right shows this relationship for BBB disruption measured in the penumbra. BBB = blood–brain barrier; WMH = white matter hyperintensity. [Color figure can be viewed at www.annalsofneurology.org]

When looking at BBB disruption in the penumbra, a similar relationship was seen where greater BBB disruption was associated with more progression of WMH (ß = 0.81, CI = 0.10–1.53, r ² = 0.10, p = 0.027) indicating that for each 1% increase in penumbral BBB disruption at baseline, there was a 0.8% increase in WMH volume over the subsequent year. This relationship is shown in Figure 3B. Baseline WMH volume also correlated with change in WMH volume (ß = 0.099, CI = 0.016–0.183, r ² = 0.11, p = 0.021).

Unadjusted Analysis: Covariates of Change in WMH Volume

The univariate model results are shown in Table 2. Of the covariates tested, only interval ischemic event during the follow‐up (ß = 0.225, CI = 0.012–0.439, p = 0.039) and scanner site (ß = −0.285, CI = −0.532 to –0.038, p = 0.025) showed a statistically significant association with change in WMH volume over 1 year in unadjusted analysis and were thus included in the multivariable model. We did not observe a correlation between change in WMH volume and clinical factors of age or history of hypertension; participants without hypertension in the sample were older (77 vs 68 years) on average.

Table 2.

Associations Between Change in WMH Volume and Predictors/Covariates in the Univariate Linear Regression Analysis

Predictor	ß Coefficient	95% CI	R 2	P
BBB in WMH	0.9484	0.3880 to 1.5080	0.1934	0.001
BBB in penumbra	0.8149	0.0981 to 1.5310	0.0982	0.027
Baseline WMH volume	0.0991	0.0155 to 0.1828	0.1060	0.021
Covariates
Age	0.0068	−0.0037 to 0.0172	0.0338	0.201
Sex	0.0503	−0.1405 to 0.2413	0.0058	0.598
Hypertension	−0.0411	−0.3159 to 0.2336	0.0019	0.765
Hyperlipidemia	0.1426	−0.0924 to 0.3775	0.0301	0.228
Diabetes	0.0225	−0.1789 to 0.2238	0.0010	0.824
Atrial fibrillation	−0.0521	−0.2598 to 0.1556	0.0053	0.616
Coronary artery disease	−0.0380	−0.2980 to 0.2221	0.0018	0.770
Tobacco use	−0.0106	−0.3286 to 0.3075	0.0001	0.947
Obstructive sleep apnea	0.0297	−0.1716 to 0.2309	0.0018	0.768
MRI site	−0.2847	−0.5315 to −0.0379	0.1008	0.025
Time since stroke	0.00129	−0.0003 to 0.0005	0.0069	0.565
Time between MRIs	−0.0009	−0.0063 to 0.0044	0.0026	0.723
NIHSS at time of stroke	−0.0144	−0.0433 to 0.0145	0.0211	0.320
Interval ischemic event	0.2254	0.0117 to 0.4390	0.0857	0.039

BBB = blood–brain barrier disruption; CI = confidence interval; MRI = magnetic resonance imaging; NIHSS = National Institutes of Health Stroke Scale; WMH = white matter hyperintensity.

Multivariable Analysis

Multivariable GLM regression was conducted that included scanner site, interval ischemic event, and BBB disruption in the WMH. Using this model, BBB disruption in the WMH (ß = 0.834, CI = 0.299–1.37, p = 0.003) and scanner site (ß = −0.224, CI = −0.448 to –0.001, p = 0.049) but not interval ischemic event (ß = 0.166, CI = −0.025 to 0.358, p = 0.087) were found to be independently associated with change in WMH volume.

A second multivariable model was evaluated that included scanner site, interval ischemic event, and BBB disruption in the penumbra as predictors. Using this model, only BBB disruption in the penumbra (ß = 0.695, CI = 0.002–1.39, p = 0.049) and scanner site (ß = −0.262, CI = −0.497 to −0.028, p = 0.029) and not interval ischemic event (ß = 0.153, CI = −0.053 to 0.360, p = 0.142) were found to be significant.

All collected clinical measures were comparable between the sites except age (mean = 64 vs 70 years), although the age difference did not reach statistical significance. Further adjusting for age and hypertension, variables known to be associated with WMH progression, in the multivariable GLM models described above did not change the results. The measured change in WMH volume was significantly different between sites (ß = −0.285, CI = −0.532 to −0.038, p = 0.025) such that the site with the younger population and smaller sample size (n = 8) showed less progression. However, when the entire analysis was repeated using only the site with the largest sample size (n = 42) the results were essentially unchanged (see Supplementary Data for details).

Additional GLM regression models were also tested that included individual BBB measures and baseline WMH volume to see if they would be independent predictors of change in WMH volume. In the model with BBB disruption in the WMH and baseline WMH volume, only BBB disruption in the WMH (ß = 0.926, CI = 0.11–1.73, p = 0.028) and not baseline WMH volume (ß = 0.006, CI = −0.11 to 0.12, p = 0.911) was associated with the outcome of change in WMH volume. However, when testing BBB disruption in the penumbra and baseline WMH volume together, neither BBB disruption in the penumbra (ß = 0.477, CI = −0.420 to 1.37, p = 0.29) nor baseline WMH volume (ß = 0.065, CI = −0.040 to 0.170, p = 0.22) were independently associated with change in WMH volume.

Prediction of WMH Progression

Change in WMH volume was dichotomized as progression (> 0), which occurred in 34 patients, or no progression (< = 0), which occurred in 16 patients. In those that progressed, the mean change in WMH volume was 0.23% (SD = 0.33). Comparing BBB disruption, expressed in 0.1% increments (which is approximately half of the average BBB disruption measured at baseline in the study) using logistic regression, found that BBB disruption measured in the penumbra (OR = 2.00, CI = 1.01–3.96, p = 0.046) was a significant predictor of WMH progression, whereas BBB disruption measured in the WMH did not reach significance (OR = 2.03, CI = 0.98–4.19, p = 0.057). This indicates that for every 0.1% increase in BBB disruption measured in the penumbra at baseline, the odds of WMH progression over the subsequent year doubled. Baseline WMH volume was not a significant predictor for discriminating participants with WMH progression (OR = 2.35, CI = 0.95–5.83, p = 0.065).

The ROC analysis found that BBB disruption measured in the penumbra was the best predictor of progression with an AUC of the ROC curve of 0.73 compared with an AUC of 0.69 for BBB disruption in the WMH and an AUC of 0.66 for baseline WMH volume. However, the AUC for BBB disruption in the penumbra was not significantly higher than BBB in the WMH (p = 0.42) or baseline WMH volume (p = 0.46). The ROC curves for all 3 are shown in Figure 4.

Figure 4.

The ROC curves for prediction of WMH progression are shown for baseline WMH volume and BBB disruption measured in the WMH and penumbra. AUC = area under the ROC curve; BBB = blood–brain barrier; ROC = receiver‐operator characteristic; WMH = white matter hyperintensity. [Color figure can be viewed at www.annalsofneurology.org]

In a sensitivity analysis using a higher threshold for WMH progression (WMH change > 0.15), both higher BBB disruption in the WMH (OR = 3.11, CI = 1.44–6.75, p = 0.004) and in the penumbra (OR = 1.87, CI = 1.12–3.14, p = 0.017) were associated with WMH progression. However, BBB disruption in the WMH had a higher odds ratio for progression and performed better on ROC analysis (AUC for WMH = 0.80 vs penumbra = 0.74). The difference in ROC performances was not significant (p = 0.21).

Discussion

This prospective observational study testing the hypothesis that BBB disruption would predict disease progression of CSVD measured as WMH volume was positive for its primary outcome. Specifically, BBB disruption measured using DSC MRI at a single timepoint in the penumbral region of the NAWM was associated with progression of WMH over the following year in patients with a history of stroke or TIA. Although prior studies have linked BBB disruption to CSVD using cross‐sectional data, ²³ , ³³ this is the first study to demonstrate the relationship between BBB disruption and WMH progression using longitudinal data. These results strongly support BBB disruption as a key driver in the pathogenesis of CSVD progression. Although it appears that BBB disruption is a biomarker for disease activity, it remains to be seen if treatments targeting BBB disruption will have an impact on disease progression.

It has been demonstrated that after a stroke, separate from any acute decrease in cognition directly related to the stroke, there is a chronic change in cognitive trajectory that hastens progression to dementia. ³⁴ This model of vascular dementia recognizes a progressive element in addition to the stepwise decline and has been characterized as infarct‐induced neurodegeneration. ³⁵ In this model, the acute injury results in a prolonged smoldering innate inflammatory response, which ultimately results in neuronal cell loss. Given that progression of WMH is associated with cognitive decline, and assuming that the BBB disruption we are measuring in the current study is a marker of inflammation, our findings are in line with this proposed mechanism of post‐stroke cognitive decline.

We found that BBB disruption measured in the WMH had the strongest association with how much the lesions would grow over the subsequent year. One way to conceptualize this would be that BBB disruption measured in the WMH represents the momentum of the disease. On the other hand, BBB disruption measured in the penumbra, which, on average, was higher than in the WMH, was found to be the better predictor of who would progress. This suggests that even in patients with minimal baseline WMH volume and WMH BBB disruption, if there is BBB disruption in the penumbra, the WMH lesions are more likely to progress. It appears that BBB disruption in the WMH provides an estimate for the rate of progression whereas BBB disruption in the penumbra is a better estimate for the risk of any progression.

However, the best threshold to define progression is not clear, ³⁶ and our primary approach of using a threshold of “any progression” is susceptible to measurement error. When using a higher threshold for progression, BBB in the WMH was a better predictor of progression, a finding that is also consistent with BBB in the WMH providing a better estimate for the expected amount of WMH progression. Longer term studies are needed to confirm if the relationship between BBB disruption in the penumbra and small changes in WMH progression are clinically the most relevant.

In unadjusted analysis, we did not identify a correlation between change in WMH volume and clinical factors of age or history of hypertension. This was somewhat unexpected given the known association of these factors with WMH progression. Possibly, this is due to the relatively young age (69 years) of the study participants and relatively high prevalence of hypertension (86%) in the setting of a modest sample size, where participants without hypertension in our sample happened to be older (on average = 77 years of age). When age and hypertension were added to the models, the results were unchanged (data not shown). We did find differences in outcome based on study site, which could be related to local population differences such as age, vascular risk factors, stroke subtypes, or access to healthcare resources. Looking more closely at this, the study site with more WMH progression was, on average, older (70 vs 64 years old); however, this difference was not significant. The other measured clinical variables were similar.

Although the focus of this study was on predicting expansion of WMH lesions, 32% of participants demonstrated some degree of decrease in WMH volume over 1 year, a proportion that is similar to prior reports of WMH regression in patients after minor stroke. ³⁷ Other studies have found WMH volumes to be dynamic with initial reductions in volume later progressing, on average, to increases in volume. ³⁸ , ³⁹ The conspicuity and volumetric measurement of WMH on FLAIR MRI may relate to underlying volume status, local inflammation, or glymphatic clearance of fluid in the interstitial space. The relatively short duration of our study limits our ability to draw conclusions about WMH regression, but future, longer‐term studies may look at these changes more critically, including how the BBB disruption may be informative in this phenomenon.

One potential criticism of this study is the use of DSC imaging to measure BBB disruption. A consensus statement about measuring BBB disruption in CSVD found disentangling the perfusion and the leakage effects on DSC imaging to be challenging. ⁴⁰ Another study aimed at assessing the feasibility of measuring BBB disruption in CSVD with DSC recommended against it, again citing difficulty separating the T1 and T2* effects on signal change. ⁴¹ Fortunately, based on a series of studies, we developed a method to separate these signals using an ATC. ³¹ Figure 1 demonstrates how, once an ATC has been applied, the effect of BBB disruption on the signal can be visualized. This pulling down of the curve is the result of a T1 effect that is caused by, and proportional to, the concentration of gadolinium that has leaked through the BBB into the brain parenchyma. It is important to recognize that this DSC approach primarily detects contrast leakage that occurs during the first (and to a certain extent the second) pass of the bolus through the brain. It will not detect delayed contrast leakage that becomes evident once the bolus enters steady state. Early and late crossing of the BBB may have different pathologic implications; thus, the results of this study should only be considered to reflect early extravasation of contrast through the BBB.

Dynamic contrast enhanced (DCE) MRI is a method that measures BBB permeability during steady state and is more appropriate for detecting delayed crossing of the BBB. ⁴² DCE imaging is known to be unreliable for measuring BBB permeability in low flow states, ⁴³ such as are seen in the WMH, but it has been used to measure BBB disruption in CSVD. ³³ DCE also requires a long image acquisition time, typically 10 to 30 minutes because it takes several minutes to reach a relative steady state. Furthermore, DCE is not commonly available on clinical scanners. DSC on the other hand is fast (< 2 minutes) and readily available on clinical scanners. Methods using diffusion‐prepared arterial spin labeling to measure BBB permeability to water have also been studied. ⁴⁴ The MRIs used in this study were research scans, but they were collected on clinical scanners using a clinical sequence that is commonly acquired in the management of stroke. DSC has been used by other groups to study the BBB disruption of white matter injury in patients after stroke. ⁴⁵ The DSC method described here has been replicated and validated in animals ⁴⁶ and independently implemented on a third‐party software platform to study stroke‐related BBB disruption. ⁴⁷ Thus, DSC is emerging as an important tool for studying the BBB.

Whereas it is well‐established that stroke is a risk factor for developing dementia, the mechanism by which this occurs remains unclear. The progressive accumulation of WMH is associated with cognitive decline ⁴ and is accelerated after stroke. ³⁸ Here, we found that disruption of the BBB in areas not yet part of the WMH was associated with WMH progression indicating that it may be a biomarker of disease activity. In practical terms, these findings suggest that we may be able to identify patients at risk of developing VCID by imaging the BBB. In addition to prognostication, this may also be helpful when designing clinical trials to test interventions to prevent VCID.

Strengths of this study include the prospective, longitudinal design and regional analysis of BBB disruption. The inclusion criteria, which limited the population to patients with post‐stroke who were without dementia, ambulatory, and verbal, resulted in a population that consisted of primarily minor stroke. Thus, it is not known if these findings would generalize to all patients with stroke. We did detect differences in the primary outcome based on the site of the MRI scan (which also reflected the site of recruitment for most patients); this may reflect differences (such as age) in the local populations served by the 2 hospitals, but there were also differences in scan parameters between the sites that could have contributed. These differences did not impact the primary findings of the study, and when the analysis was repeated using a single site, the findings were unchanged (see Supplementary Data). We calculated change in WMH volume relative to brain volume instead of intracranial volume. This was done to minimize the effects of brain atrophy on the measurements which can be substantial after stroke, but the resulting change in volume reflects a relative change in WMH burden, not an absolute change in volume. To minimize the effects of encephalomalacia of the prior stroke on our measurements of WMH volume, the first MRI timepoint was at least 3 months post‐stroke, and the analysis was limited to the contralateral hemisphere when possible. Whereas brain volume loss is most rapid within 3 months of stroke, ⁴⁸ atrophy can continue beyond that time frame and lead to underestimation of WMH progression in patients with bilateral strokes. However, in our study, participants with bilateral strokes had only small and scattered infarcts that are less likely to have a substantial impact on volume measurements. Another limitation of the study is the modest sample size which limited our ability to include additional predictors of WMH progression. The poor retention rate (50/81), which was largely due to the coronavirus disease (COVID) pandemic, is also a potential limitation. The definition of WMH progression as any increase in volume is susceptible to measurement error but should not introduce a bias in favor of the prespecified hypothesis, and our sensitivity analysis using a higher threshold for progression supported the primary results. The use of DSC to measure BBB disruption may be considered a limitation by some, but provided the results are interpreted with an understanding of the biomarker being used, the findings have important implications. Although the slice thickness of our DSC acquisition limited our analysis to in‐plane rather than 3D, the use of a short, clinically available acquisition will facilitate clinical translation of BBB disruption as a biomarker. Finally, the changes in WMH detected in this study were over a relatively short duration of 1 year; a longer study to validate these findings is needed.

Conclusions

In our cohort of patients with chronic cerebrovascular disease, BBB disruption measured with DSC MRI provided important prognostic implications for future disease progression. The location and severity of BBB disruption detected was informative about disease pathogenesis and activity. BBB imaging may provide a window into brain health not previously recognized.

Author Contributions

R.L. and C.B.W. contributed to conception and design of the study. R.L., K.C.K., N.Y.W., R.F.G., and C.B.W. contributed to the acquisition and analysis of data. R.L., K.C.K., and N.Y.W. contributed to the drafting the text or preparing the figures.

Potential Conflicts of Interest

C.B.W. reports honoraria from Uptodate.com for articles written on vascular dementia. The other authors report no disclosures.

Supporting information

Supplementary Table S1. Participant characteristics by site.

Data Availability

Upon reasonable request to the corresponding author, the data will be made available after close‐out and publication of the parent study under a formal data sharing agreement and with approval from the requesting researcher's local ethics committee.