Neuroimaging biomarkers of disease progression and cognitive change in patients with retinal vasculopathy with cerebral leukoencephalopathy

Abstract

Background and Objectives:

A monogenic, age-related cerebral small vessel disease (cSVD), retinal vasculopathy with cerebral leukoencephalopathy and systemic manifestations (RVCL-S), causes accelerated vascular dementia, vision loss, and premature death. Given knowledge is limited regarding its pathophysiology, we examined the natural history of progression in imaging and cognitive endpoints to define rate and variability of progression in a prospective RVCL-S cohort. We hypothesized that cerebral blood flow (CBF) and oxygen extraction fraction (OEF), as metrics of cerebral hypoxia-ischemia, would be associated with neurological disease progression.

Methods:

We performed sequential brain MRI and cognitive battery in a prospective RVCL-S cohort. Arterial spin labeling and asymmetric spin echo quantified CBF and OEF in normal-appearing white matter (WM), respectively. Three neuroimaging endpoints of cSVD included: log-transformed, normalized FLAIR white matter hyperintensity (WMH) volume, WM microstructure using mean diffusivity, and normalized WM volume. Five cognitive and motor endpoints included: Digit Symbol Substitution Test (DSST), Category Fluency, Free Recall, Montreal Cognitive Assessment (MoCA), and Gait Speed. Linear mixed-effects models examined age, CBF, and OEF in association with neuroimaging and cognitive progression.

Results:

Twenty-five participants, aged 23–68 years (median 47 years), 56% female, underwent 151 scans over median [25^th, 75^th] of 2.2 [1.8, 4.4] years. All neuroimaging and cognitive endpoints progressed over time except for MoCA. Of neuroimaging endpoints, WMH volume demonstrated the largest rate of change (+31.3%/year, 95% CI: 20.8%, 42.3%). Of cognitive endpoints, DSST, a metric of processing speed, showed the steepest decline (−13.1 T-score points/decade, 95% CI: −21.2, −5.32), followed by Free Recall and Category Fluency: −5 [−7.8, −2.4] and −4.1 [−7.2, −1.2] T-score points/decade, respectively. Age at first brain lesion was estimated to be 30.6 [23.8, 35.8] years modelled from individual WMH volume trajectories. In multivariable analysis, OEF, but not CBF, was independently associated with WMH growth (β=3.73, 95% CI 0.63–6.83, P=0.029) and change in WM microstructure (β=0.175, 95% CI 0.029–0.32, P=0.029). Neither OEF nor CBF was independently associated with cognitive decline.

Discussion:

Elevated OEF in at-risk WM was associated with progression in WMH and impairment in WM microstructure, suggesting a role for tissue hypoxia-ischemia underlying RVCL-S pathophysiology. Cerebral OEF holds promise as a predictive biomarker to risk-stratify patients with RVCL-S and other forms of cSVD.

Introduction

Retinal vasculopathy with cerebral leukoencephalopathy and systemic manifestations (RVCL-S) is a rare, autosomal dominant microvasculopathy presenting in early to mid-adulthood with progressive vision loss and vascular contributions to cognitive impairment and dementia (VCID). RVCL-S is caused by a frameshift mutation of the carboxyl-terminus of three prime exonuclease-1 (TREX1), the major human 3’−5’ exonuclease on chromosome 3p21.1-p21.3.^1,2 Individuals with RVCL-S develop characteristic neuroimaging findings including: (1) white matter hyperintensities (WMH) on MRI, which commonly show restricted diffusion and nodular contrast enhancement, (2) periventricular, rim-enhancing pseudotumor lesions, and (3) white matter (WM) calcifications on computed tomography.^3,4 Observational studies have shown WM atrophy, disruption in microstructural integrity, and reduction in cerebrovascular reactivity (CVR).^5,6 RVCL-S affects the microvasculature systemically and may cause anemia, nephropathy, liver disease, and Raynaud’s phenomenon beginning as early as the 20s.^1,7,8 Individuals experience progressive disability due to loss of vision, migraines, anxiety, and gait imbalance, followed by dementia, with death due to sequelae of VCID in the 40’s to 60’s.^4,5,9,10 As RVCL-S is rare and under-recognized,^4,8 it has been challenging to define its natural history which has limited clinical trial preparedness, including the ability to define age of disease onset.

It is unclear how mutations in TREX1 result in vascular dementia, but neuropathologic studies implicate chronic ischemia via arteriolar wall thickening, luminal narrowing and vessel destruction.⁶ In sporadic cSVD, imaging biomarkers of microvascular ischemia, including decreased cerebral blood flow (CBF) and impaired CVR, predict progression of WMH and microstructural loss.^11–13 Similar to CBF and CVR, we and others have hypothesized that elevated oxygen extraction fraction (OEF) may serve as a marker of active hypoxia-ischemia,^14,15 having been predictive of future stroke in extracranial carotid occlusive disease.^16,17 Whether OEF predicts progression in cSVD and VCID, however, has not been established. We have previously shown in patients with sporadic cSVD and sickle cell disease, that OEF is associated with WMH burden and loss of microstructural integrity.^12,18,19 In a cross-sectional RVCL-S cohort, we found that OEF increased across age compared to aging controls.⁵ Beyond RVCL-S, developing biomarkers for VCID has been of interest, especially biomarkers implicated in disease pathogenesis.^20–22 Such biomarkers hold promise to aid in selecting individuals at greatest risk who may benefit from early intervention.^23,24 Given neuroimaging manifestations and dementia progress quickly, RVCL-S may serve as an accelerated model for sporadic cSVD associated with aging and vascular risk factors.

In this prospective study, we had two major goals: (1) to define rate and variability of progression in neuroimaging and cognitive endpoints to guide endpoint selection for clinical trials, and (2) to examine imaging biomarkers associated with cerebral and cognitive progression to identify individuals at risk for early onset and/or severe disease. We hypothesized that MRI measures of hypoxic-ischemic physiology, using CBF and OEF, would be associated with progression in neuroimaging and cognitive endpoints, thus implicating the critical role of chronic hypoxia-ischemia as a pathomechanism in RVCL-S.

Materials and Methods

Standard Protocol Approvals, Registrations, Patient Consents, and Study Participants

This study was approved by the Human Research Protection Office at the Washington University School of Medicine. All participants provided written informed consent. This study included a single-center, prospective cohort of adults with genetically confirmed RVCL-S. Participants had study visits including a MRI scan and cognitive assessment performed clinically as part of routine surveillance at baseline, every 6–12 months, and additionally if participants had neurological symptoms warranting further evaluation. RVCL-S participants were required to have two or more study visits at least six months apart, to be included, permitting a temporal association analysis of disease progression. Exclusion criteria were contraindications to MRI, infection or illness within 4 weeks, and pregnancy, while other comorbidities were permitted. Systemic disease features of RVCL-S were assessed by medical history and chart review and are defined in the eMethods. Common neuroradiologic features were present if described in the radiology report. A prospective “referent” cohort was enrolled specifically for estimating age of first lesion development (eMethods). At each study visit, brain MRI, cognitive assessment, vital signs, laboratories, medications, and interim illnesses were obtained (eFigure 1). Abbreviations are in eTable 1.

Cognitive, Depression, and Anxiety Assessment

During each study visit, a standardized cognitive battery was performed including five tests: Digit Symbol Substitution test (DSST)^25,26 for working memory and processing speed; category fluency^27–29 (animal and vegetable naming); the Buschke and Grober Free and Cued Selective Reminding Test with Immediate Recall (Free Recall)^30,31 for episodic memory; Montreal Cognitive Assessment (MoCA);³² and, Gait Speed,³³ a measure of cognitive and motor processing. Depression and anxiety levels were assessed using the Geriatric Depression Scale³⁴ and the State-Trait Anxiety Inventory,³⁵ respectively. Raw test scores were converted to T-scores derived from the study sample permitting ability to track changes over time. Cognitive testing began one year after study start, thus,139 cognitive assessments were available for 151 study visits.

Imaging Protocol

At each study visit, participants underwent a brain MRI on a 3 Tesla scanner (Siemens Healthineers, Erlangen, Germany) with a 64-channel head coil. Standard T1-weighted Magnetization-Prepared-RApid-Gradient-Echo (MPRAGE); 2D FLuid-Attenuated Inversion Recovery (FLAIR); We utilized a pseudocontinuous arterial spin labeling (pCASL) to measure voxel-wise CBF.³⁶ Blood T1 required for CBF quantification.³⁷ An asymmetric spin echo sequence was obtained to measure voxel-wise OEF.^38,39 For quantifying WM microstructure, diffusion weighted imaging was acquired using echo planar imaging with 25 non-collinear diffusion directions. See eMethods for detailed imaging protocol.

Image Processing

For tissue-type segmentation, MPRAGE images were skull-stripped and segmented into gray matter (GM) and WM using Statistical Parametric Mapping (SPM12, Wellcome Institute of Neurology, London, UK).⁴⁰ Using FMRIB Software Library (FSL, Oxford University, UK), FMRIB’s Integrated Registration and Segmentation Tool (FIRST) segmented subcortical nuclei.⁴¹ FMRIB’s Linear (rigid) Image Registration Tool (FLIRT) registered images obtained within a scan session.^42,43 Accuracy of image alignment was confirmed by visual inspection. Before partial volume correction, all ASE and pCASL images were performed with motion correction using FSL MCFLIRT and frames affected by motion were excluded.

To segment and quantify WMH volume on each scan, an automated segmentation approach was used.⁴⁴ Subsequently, the WMH mask was manually edited by a trained student (V.C.) and further reviewed by a board-certified vascular neurologist with 19 years of experience (A.F.) to ensure accurate lesion voxels were captured by the automated segmentation. Readers were blinded to patient identifiers and study visit date. Normal-appearing WM (NAWM) region was obtained by subtracting the WMH mask from the segmented WM mask. To account for atrophy of brain over time, WMH volumes were adjusted by total brain volume (GM+WM volumes), defined as “normalized WMH volume”.^45–47 Diffusion tensor parametric maps of mean diffusivity (MD) were calculated using FSL diffusion toolbox (FDT, Oxford University, UK).⁴⁸ Three of 151 study visits did not acquire diffusion tensor imaging due to patient request for shorter scan duration and thus were missing data. For quantification of WM atrophy, the MPRAGE was segmented by tissue-type. To account for inter-individual variation in head size, WM volume was normalized to total intracranial volume (GM+WM+cerebrospinal fluid volumes), defined as “normalized WM volume”.⁴⁹ As putative neuroimaging predictors of progression of cSVD, individual mean NAWM CBF and OEF were calculated as previously described (eMethods).⁵⁰ Mean NAWM OEF was normalized to whole brain OEF to obtain a WM-specific measure of OEF (nOEF) that minimizes global OEF variation, such as due to anemia or age, across patients. One participant did not have pCASL performed causing one missing CBF in the overall dataset.

Statistical analysis

Statistical analyses were performed with R version 4.4.0 and SAS version 8.3 (Cary, NC). Two-sided alpha was used. To adjust for false discovery rate (FDR) due to multiple tests, P-values were adjusted using the Benjamini-Hochberg procedure and considered statistically significant at <0.05 (see eMethods for details). The statistical analysis included a sensitivity analysis and two primary study analyses. The sensitivity analysis considered potential confounding effects of two study medications administered to RVCL-S participants during the study follow-up period (included in eMethods). For all models tested for the two main study analyses, three neuroimaging endpoints included: (1) log-transformed normalized FLAIR WMH volume with increases indicating disease progression, (2) microstructural integrity defined as WM MD with increases indicating disease progression, and (3) normalized WM volume with decreases indicating progression in WM atrophy. These imaging endpoints were chosen based on prior studies of RVCL-S and sporadic cSVD demonstrating greater change in WM injury and volume metrics than GM.⁴⁷ The five cognitive endpoints included DSST, Category Fluency, Free Recall, Gait Speed, and MoCA, converted to T-scores standardized to the RVCL-S cohort using cohort-specific mean and standard deviations to enable within-cohort comparisons of relative performance over time. The first analysis was a natural history study to define the rate of change and variability in neuroimaging and cognitive endpoints in the RVCL-S cohort. Age, as a proxy for disease duration, was examined as a predictor of each endpoint using linear mixed-effects models (lme4 package in R) with age at time of scan as a predictor variable (fixed effect), with participant included as a random effect with random intercepts but fixed slopes for time (age). Details are included in the eMethods. For estimating age and variability of first cerebral lesion development in RVCL-S, subject-level linear associations of WMH volume progression estimated the age at which the modelled progression line intersected with the reference WMH volume line from the referent cohort with four outliers removed (see eMethods). The second analysis examined biomarkers of neuroimaging and cognitive progression using unadjusted and adjusted linear and linear mixed-effects models. The second analysis was divided into three sub-analyses: (a) examination of NAWM CBF and nOEF in association with change in neuroimaging endpoints; (b) examination of NAWM CBF and nOEF in association with change in cognitive endpoints, and (c) examination of neuroimaging endpoints in association with change in cognitive endpoints. In the adjusted models, age at time of scan, sex, and time between study visits were entered as covariates. To adjust for existing structural or cognitive impairment which is known to predict subsequent test results, imaging (or cognitive) endpoint value at current examination was also included. For linear mixed-effects models, subject was included as a random effect.

Data availability statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

Results

Cohort characteristics

Between 5/16/2016 to 9/27/2023, 30 participants with genetically-confirmed RVCL-S underwent 156 study visits. Five participants had a single study visit and were excluded from analytical dataset, while the remaining included individuals (n=25) had two or more study time-points at least six months apart and five died prior to data lock. Of the five participants with a single study visit, one died before the second scan, while four others had difficulty with the commute and sought care locally.

Table 1 shows RVCL-S baseline characteristics. Median age at first study visit was 47 [38, 53] years (range 23–68). Fourteen (56%) participants were female. Systemic comorbidities in RVCL-S are in Table 1. Additionally, three of 25 (12%) patients had a diagnosis of diabetes mellitus. Three (12%) were active tobacco users. One (4%) had atrial fibrillation and took apixaban. Sixteen (64%) were taking statin medications and 10 (40%) were taking antiplatelet medications (aspirin or clopidogrel) to maximize stroke prevention. RVCL-S participants underwent 151 MRI scans with a median of 5 [4, 8] scans over 2.2 [1.8, 4.4] years and a scan interval of 6.1 [5.6, 8.4] months. A referent cohort (n=42, aged 25–65) underwent a single study visit and provided a healthy cohort for comparing baseline neuroimaging and cognitive endpoints (eTable 2). As cognitive assessment was added after study initiation, 12 of the 151 RVCL-S study visits and 11 of 42 referent study visits were missing cognitive assessment.

Table 1.

Baseline characteristics of RVCL-S cohort

		RVCL-S N=25
Age at baseline scan (years)		47 [38, 53]
Age at last scan (years)		50 [43, 57]
Age at death (years) (N=6)		54 [48, 60]
Female sex		14 (56%)
White race		24 (96%)
Number of years of education		16 [13, 16]
Number of scans		5 [4, 8]
Interval between scans (months)		6.1 [5.6, 8.4]
Total duration between baseline and last scan (years)		2.2 [1.8, 4.4]
Common disease features a
Vision disturbance		22 (88%)
Focal neurological deficits		12 (48%)
Cognitive impairment, per patient or family member		14 (56%)
Psychiatric disturbance		19 (76%)
Seizures		3 (12%)
Migraine headache		15 (60%)
History of malignancy		3 (12%)
Family history of malignancy		13 (52%)
Common neuroradiological features b
Punctate white matter lesions on FLAIR		24 (96%)
Pseudotumor ring-enhancing lesions with vasogenic edema on FLAIR		8 (32%)
Contrast-enhancing lesions		19 (76%)
Diffusion weighted imaging (DWI) restricted lesions		13 (52%)
Other systemic manifestations c
Retinopathy		22 (88%)
Liver disease		10 (40%)
Anemia		20 (80%)
Nephropathy		18 (72%)
Hypertension requiring medications		17 (68%)
Raynaud’s		14 (56%)
Neuroimaging endpoints (baseline value shown)
Normalized WMH volumed		0.005 [0.001, 0.014]
White matter MD × 10−3 (mm2/s)		0.697 [0.685, 0.710]
Normalized white matter volume		0.291 [0.277, 0.307]
Cognitive endpoints (baseline value shown)e	Raw Values	T-scores
DSST (number per 90 seconds)	38 [31, 48]	51.2 [46.6, 57.8]
Category fluency (words per minute)	29 [27, 32]	50.2 [47.6, 54.0]
Free recall (total of three trials)	32 [24, 35]	52.3 [42.8, 55.9]
MoCA (total score of 30 maximum)	27 [25, 28]	53.6 [47.9, 56.5]
Gait speed (meters/second)	1.2 [1.0, 1.2]	53.5 [48.3, 54.7]

Data are shown as median [25^th, 75^th].

^aCommon disease features were present if documented within the medical record by the treating physician. Malignancy history included one individual with a history of skin cancer, one with a history of breast cancer, and one with a history of breast and bladder cancer. Focal neurological deficits included any report of language, visual-spatial, motor, sensory, cranial nerve, coordination, or balance deficits in the chart, except that any vision deficits were included in the Vision Disturbance category.

^bCommon neuroradiological features were present if described in the radiology report or identified on image review by a board-certified vascular neurologist (A.F.).

^cOther systemic manifestations were present as follows: liver disease if alanine transaminase, aspartate transaminase, or alkaline phosphatase were elevated beyond the laboratory’s reference range, anemia if either hemoglobin or hematocrit were low, nephropathy if either serum creatinine or urine protein were elevated, hypertension if the patient required daily home antihypertensive medications, and Raynaud’s if listed as a clinical diagnosis.

^dNormalized WMH volume is unitless as it is the volume of white matter hyperintensity on FLAIR in ml divided by the sum of the gray and white matter volumes in ml. This normalization to whole brain was performed to account for brain atrophy over time.

^eRaw scores and T-scores are shown. The T-scores were derived using the mean and pooled standard deviation from the RVCL-S cohort to facilitate within-cohort standardization and to track relative changes over time.

WMH=White matter hyperintensity; White matter MD=mean diffusivity (mm²/s); DSST=Digit Symbol Substitution Test; MoCA=Montreal Cognitive Assessment.

Natural history of neuroimaging and cognitive progression in RVCL-S

Longitudinal changes in neuroimaging and cognitive endpoints as a function of age, as a proxy for disease duration, are shown in Figure 1, with independent estimates for age at time of scan shown in Table 2. Plots of the raw data for imaging and cognitive endpoints with a reference line to the referent cohort are shown in eFigure 2. Age was associated with progression in all three neuroimaging endpoints (Table 2). Normalized WMH volume increased within individuals at 0.27 units per year (95% CI 0.18, 0.35, P=0.0001) or 2.7 units per decade (95% CI: 1.8, 3.5) (Figure 1A). This corresponds to an increase of 31.3% per year (95% CI: 20.8%, 42.3%) in raw, non-transformed WMH volume data. WM MD, as an indicator of decline in microstructural integrity, revealed that WM MD appeared stable during younger adult ages, however increased exponentially with age, with the upturn near age 40 (β=0.0002 (SE=0.0001), 95% CI: −0.0001, 0.0003, P=0.0001) (Figure 1B). Normalized WM volume, as an indicator of atrophy, decreased within patients at a rate of −0.012 per decade (~15cc of tissue loss, 95% CI: −0.008, −0.001, P=0.0001) (Figure 1C).

Figure 1. Natural history of neuroimaging and cognitive changes in RVCL-S.

Using age as a proxy for disease duration as the main predictor in a linear mixed effects model, longitudinal changes in neuroimaging and cognitive endpoints in RVCL-S are plotted for: (A) log-transformed normalized white matter hyperintensity (WMH) volume, (B) white matter (WM) mean diffusivity (MD) (x 10⁻³ mm²/s), (C) normalized WM volume, (D) digit symbol substitution test (DSST) t-score, (E) Category Fluency t-score, (F) Free Recall t-score, (G) Montreal Cognitive Assessment (MoCA) t-score, and (H) Gait Speed t-score. Raw P-values for age as a predictor are shown. Age remained significantly associated with progression in all endpoints except for MoCA before and after Benjamin-Hochberg adjustment for multiple testing. Dots of the same color or symbol represent a single patient.

Table 2.

Longitudinal change in neuroimaging and cognitive endpoints in RVCL-S

Main effect: Agea	βb	SE	Conditional R2	Residual Variance	P-valuec	95th % CI
						Lower	Upper
Neuroimaging Endpoints
Log normalized WMH Volumed	0.274	0.039	0.994	0.081	0.0001	0.189	0.353
White matter MD	−0.013	0.0038	0.951	0.0002	0.001	−0.0205	−0.005
White matter MD (Age2)e	0.0002	0.0001			0.0001	−0.0001	0.0003
Normalized white matter Volumef	−0.0012	0.0003	0.889	<0.0001	0.0001	−0.008	−0.001
Cognitive Endpoints
DSST T-score	−1.33	0.402	0.988	8.71	0.01	−2.12	−0.532
Category Fluency T-score	−0.412	0.15	0.738	29.73	0.015	−0.719	−0.117
Free Recall T-score	−0.501	0.135	0.605	43.27	0.005	−0.776	−0.237
Gait Speed T-score	−0.345	0.1587	0.713	31.4	0.046	−0.688	−0.0244
MoCA T-score	−0.251	0.144	0.438	32.34	0.092	−0.569	0.0395

^aAge was used as a proxy for disease duration to assess change in imaging and cognitive measures over time. Each model was adjusted by participant ID as a random effect.

^bBeta estimate represents change per decade.

^cP-values adjusted using Benjamini-Hochberg Test for multiple comparisons.

^dWMH volume was normalized to whole brain volume and then log-transformed.

^eAge was squared for this model.

^fNormalized white matter volume was normalized to total intracranial volume.

β=estimate; SE=Standard error; CI=confidence interval; WMH=white matter hyperintensity; WM MD=mean diffusivity (mm²/s) as measured using diffusion tensor imaging; DSST=Digit Symbol Substitution Test; MoCA=Montreal Cognitive Assessment.

Given age of disease onset and defining a threshold at which patients become symptomatic in RVCL-S are unclear, we estimated the age of cerebral lesion development by modelling individual WMH progression lines and finding the intersection between an individual’s WMH trajectory and the referent WMH reference line (Figure 2). Age of cerebral WMH development was a median of 30.6 [23.8, 35.8] years.

Figure 2. Age at first cerebral lesion development in RVCL-S.

Age of onset for cerebral lesion development was estimated in the RVCL-S cohort. Subject-level linear predictions of white matter hyperintensity (WMH) volume progression (one color per dot/line per patient) were used to estimate the age at which the modelled progression line intersected with the reference WMH volume line from the control cohort. Age of cerebral WMH development was a median 30.6 [23.8, 35.8] years ranging from 10.8 to 48.5 years.

Of five cognitive tests, four declined significantly with increasing age within individual patients (Figure 1, Table 2). DSST, which had a median score of 52 [46, 62] in the referent cohort, showed the greatest magnitude of change with −13.3 (95% CI: −21.2, −5.3) T-score points intra-individual decline in the RVCL-S cohort per decade, with many patients showing precipitous drops (Figure 1, P=0.01, panels D-H). Free recall, followed by category fluency, showed the next steepest changes with −5 [−7.8, −2.4] (P=0.005) and −4.1 [−7.2, −1.2] (P=0.015) points per decade, respectively. Unlike the other cognitive tests, administration of category fluency does not require intact vision which is relevant in later stages of RVCL-S. Gait speed is a measure of cognitive and motor function and had low residual variance suggesting a good model fit, however did not steeply change (−3.45 [−6.9, −0.24] T-score points per decade, P=0.046). MoCA did not change significantly with age.

Cerebral hemodynamics and oxygen extraction in association with progression in neuroimaging endpoints of RVCL-S

We hypothesized that WM hypoxia-ischemia may predict WMH progression and WM structural and microstructural change in RVCL-S. As metrics of cerebral hemodynamic and oxygen metabolic stress, NAWM CBF and nOEF, respectively, were examined in association with longitudinal progression in three neuroimaging endpoints. Linear mixed-effects analysis with CBF (or nOEF) as a single fixed effect showed associations of CBF and nOEF with subsequent WM mean diffusivity and log-transformed normalized WMH volume, respectively (eTable 3). On multivariable analysis, CBF was not associated with disease progression, while nOEF was independently associated with progression in WMH volume (β=3.73, SE=1.56, 95% CI 0.63, 6.83, P=0.029) and WM microstructural change, defined by MD (β=0.175, SE=0.074, 95% CI 0.029, 0.32, P=0.029) (Table 3, eFigure 3). nOEF was not independently associated with WM atrophy, suggesting that atrophy may reflect a separate neurodegenerative or accelerated aging process in RVCL-S distinct from hypoxia-ischemia, the latter which is tied to WMH lesion progression and decline in WM microstructure. Representative patient examples demonstrating examples of varying baseline nOEF in relation to progression in FLAIR WMH expansion at 6–12 months are shown in Figure 3.

Table 3:

CBF and nOEF in association with progression in neuroimaging endpoints of cSVD

Dependent variablea	Independent variable
	CBF
	β (SE)	P- valueb	95% CI (lower, upper)
Log normalized WMH Volume	0.006 (0.006)	0.465	(−0.0058, 0.018)
Log WM Mean Diffusivity	−9.9 e-5 (0.0003)	0.740	(−0.0007, 0.0005)
Normalized WM Volume	−0.0002 (0.0001)	0.465	(−0.0005, 0.0001)
Dependent variablea	Independent Variable
	nOEF
	β (SE)	P- valueb	95% CI (lower, upper)
Log normalized WMH Volume	3.73 (1.56)	0.029	(0.63, 6.83)
Log WM Mean Diffusivity	0.175 (0.074)	0.029	(0.029, 0.32)
Normalized WM Volume	0.030 (0.04)	0.45	(−0.049, 0.11)

^aDependent variable was neuroimaging endpoint at subsequent study visit incorporating all study visits in a linear model. These models were adjusted for the following variables: age at current study visit, sex, neuroimaging endpoint value at current study visit, and time between two study visits.

^bP-values were adjusted using Benjamini-Hochberg Test for multiple comparisons.

CBF=Cerebral blood flow (ml/min/100g); β=estimate; SE=standard error; CI=confidence interval; nOEF=normalized oxygen extraction fraction (unitless); WMH=white matter hyperintensity; WM Mean Diffusivity = white matter mean diffusivity (mm²/s) as measured using diffusion tensor imaging.

Figure 3. Representative patients demonstrating varying baseline normalized oxygen extraction fraction (nOEF) maps in relation to progression in white matter hyperintensities (WMH) on FLAIR MRI.

(A) A young adult with RVCL-S demonstrated low baseline nOEF with no significant disease progression over one year, in contrast to (B) another similarly aged patient who demonstrated elevated baseline nOEF and who showed progression in a left frontal horn pseudotumor over one year (red arrows). (C) An older adult with RVCL-S demonstrated low baseline nOEF with no significant disease progression in FLAIR lesions over six months, while (D) another older adult with RVCL-S demonstrated higher nOEF with substantial progression in FLAIR hyperintensities and new areas of nodular enhancement (red arrows) in the white matter over six months.

Cerebral hemodynamics and oxygen extraction in association with cognitive decline in RVCL-S

We hypothesized that WM hypoxia-ischemia would be associated with decline in cognitive and motor performance, thus, CBF and nOEF were examined in association with longitudinal cognitive change. Linear mixed-effects analysis with a single fixed effect showed that nOEF was significantly associated with progression in DSST (β=−73.46, P=0.046) (Figure 4, eTable 3). On multivariable analysis, however, neither CBF nor nOEF remained independently associated with cognitive decline (eTable 4).

Figure 4. Linear mixed-effects analysis of CBF (A, C, E, G) and nOEF (B, D, F, H) in association with subsequent cognitive test result (T-scores).

CBF was associated with subsequent DSST and category fluency, while nOEF was associated with subsequent DSST (panel B) (β=−73.46, raw P=0.009, Benjamini-Hochberg adjusted P=0.046). Dots of the same color or symbol represent a single patient.

Neuroimaging endpoints in association with progression in cognitive decline in RVCL-S

Given RVCL-S clinical trials are likely to have a small sample size challenging the ability to see effects on cognitive decline, it will be helpful to know which neuroimaging endpoints are most associated with cognitive decline. We examined whether neuroimaging endpoints were associated with cognitive decline, and if so, which imaging metric was the strongest predictor when all three were included in the same model. On multivariable analysis, all three neuroimaging endpoints were associated with decline in one cognitive test. Log-transformed, normalized WMH volume was associated with reduced performance on DSST (P=0.023). Normalized WM volume was associated with decline in DSST (P=0.006). WM MD was associated with decline in category fluency (P=0.033). Neuroimaging endpoints did not predict progression in free recall, gait speed, or MoCA (Table 4).

Table 4.

Measures of cSVD neuroimaging metrics in association with progression in cognitive impairment in RVCL-S

Dependent variablea	Independent variables
	Log Normalized WMH Volume			Log White Matter MD			Log Normalized White Matter Volume
	β (SE)	P-valueb	95% CI (lower, upper)	β (SE)	P- valueb	95% CI (lower, upper)	β (SE)	P-valueb	95% CI (lower, upper)
DSST T-score	−2.44 (0.98)	0.022	(−4.29, −0.56)	−16.3 (24.76)	0.512	(−62.64, 33.58)	41.33 (13.25)	0.006	(16.22, 66.54)
Free Recall T-score	−1.55 (1.31)	0.240	(−3.98, 0.91)	−77.16 (36.21)	0.108	(−143.45, −1.08)	28.45 (17.82)	0.174	(−4.18, 63.8)
Category Fluency T-score	−0.65 (1.16)	0.574	(−2.82, 1.53)	−82.36 (31.9)	0.033	(−142.00, −20.35)	29.02 (16.48)	0.125	(−1.32, 62.64)
Gait Speed T-score	−2.47 (1.32)	0.099	(−4.96, 0.004)	−42.74 (36.7)	0.248	(−111.03, 0.294)	42.71 (21.02)	0.099	(3.68, 81.94)
MoCA T-score	−2.84 (1.63)	0.258	(−5.80, 0.565)	−23.43 (45.82)	0.611	(−107.17, 68.21)	25.16 (21.95)	0.386	(14.56, 69.19)

^aDependent variable was neuroimaging metric at subsequent study visit. A single model included three neuroimaging metrics of cSVD as predictors of cognitive decline to determine which would be most predictive of each cognitive outcome. Each model was adjusted by age at study visit, sex (male=1), time between two study visits, and participant ID as a random effect.

^bP-values adjusted using Benjamini-Hochberg Test for multiple comparisons.

WMH=white matter hyperintensity; WM Mean Diffusivity=white matter mean diffusivity (mm²/s) as measured using diffusion tensor imaging; DSST=Digit Symbol Substitution Test; MoCA=Montreal Cognitive Assessment; β = Beta coefficient; SE=standard error; CI=confidence interval.

Discussion

In a longitudinal study of RVCL-S patients, MRI measures of cerebral hypoxia, defined as elevated nOEF, was associated with progression in cSVD, defined as WMH growth and loss of microstructural integrity, over a median follow-up of 2.2 years. Furthermore, tissue hypoxia was associated with decline in DSST, a measure of processing speed and attention, which declined early on (40s) in the RVCL-S cohort. Our findings suggest: (1) a role for hypoxia-ischemia underlying the pathophysiology of RVCL-S, and (2) consideration of cerebral OEF as a predictive biomarker of VCID progression in RVCL-S and potentially other forms of cSVD. Another goal of this prospective study was to define the time-course of neuroimaging and cognitive trajectories in RVCL-S so that optimal clinical endpoints could be selected and used in future clinical trials. Patients with RVCL-S aged 23 to 68 were followed over 1 to 7 years with a median of 5 study visits over 2.2 years. Of the endpoints tested, all three neuroimaging and four of five cognitive tests progressed significantly in association with age, suggesting their consideration in future trials. Further examination of each endpoint suggested that log-transformed normalized WMH volume, of neuroimaging endpoints, and DSST, of cognitive endpoints, changed early and steeply in the disease, and may be optimal for use in clinical trials.

White matter OEF as a biomarker of small vessel ischemia in RVCL-S

Brain pathology in RVCL-S shows thickened, fibrous arterioles with multilaminated basement membranes and TREX1 staining in endothelial cells.^1,4,51 Markers of chronic endothelial activation (von Willebrand factor and angiopoietin-2)^10,52 and impaired CVR suggest a chronic, non-inflammatory small vessel vasculopathy.⁵³ Sequential brain MRI shows accelerated progression in cSVD (WM lesions, microstructural injury, atrophy) and signs of endothelial barrier disruption with nodular enhancement and pseudotumor formation.^5,53 Based on known pathology and radiological findings of RVCL-S, with multiple neuroimaging and cognitive features of cSVD and VCID, we hypothesized that decreases in CBF and increases in nOEF would be associated with RVCL-S disease severity and may serve as an early and potentially reversible biomarker for disease progression beyond known predictors of cSVD progression such as WMH volume and age. We found that nOEF predicted progression in two neuroimaging endpoints, normalized WMH volume and WM MD, both which are common metrics of tracking WM injury in other, more common cSVDs.^{11,13,20,47,54} Indeed, we found examples of patients at both young (late 30s) and older (mid 50s) ages who had elevated nOEF and greater WMH progression compared to patients of similar age who did not substantially progress (Figure 3). Normalized OEF was not associated with WM atrophy suggesting that atrophy may reflect mechanisms other than hypoxia-ischemia. Beyond a premature microvasculopathy, dysfunctional TREX1 may lead to accelerated brain aging due to defective DNA repair. Alternatively, the study was not adequately powered to detect a change in this endpoint.

Studies of sporadic cSVD with aging and hypertension have sought neuroimaging and fluid biomarkers that predict risk of progression in VCID.^20,21,24 Physiological neuroimaging biomarkers such as reduction in regional CBF and CVR may be helpful for identifying subjects with early disease onset or greater disease severity.^11,54 Other biomarkers, such as WMH burden, WM microstructure on DTI, or neurofilament light chain in cerebrospinal fluid, may represent later disease stages when irreversible damage may have occurred.^11,13 Tissue-based OEF has been tested primarily in the setting of hyperacute ischemic stroke as a metric of ischemic penumbra and in the setting of large artery stenosis or occlusion.^14,17,55 Increased OEF, normalized to the contralateral hemisphere, measured using PET, predicted ipsilateral ischemic stroke distal to proximal internal carotid artery occlusion. Tissue-based OEF in cSVD has been tested in cross-sectional studies of RVCL-S, sickle cell disease, and sporadic cSVD.^{5,12,18,19,50} and is increased relative to controls, however longitudinal studies measuring the predictive ability of nOEF for disease progression outside of RVCL-S have been limited.^56,57

While normalized OEF was associated with progression in cognitive domains most affected in VCID including DSST, a measure of processing speed and working memory.^21,26,58,59 it was no longer significant in multivariable analysis. This is possibly explained by a low sample size in the setting of increased intra-individual variability known to occur with cognitive testing. Indeed, detection of significant cognitive decline greatly increases sample sizes required relative to the numbers needed to detect neuroimaging progression in sporadic cSVD.⁴⁶

In contrast to OEF, measures of CBF did not show an association with cerebral disease progression. It is possible that there are competing mechanisms influencing CBF. In the setting of microvascular hypoxia-ischemia, cerebral autoregulation may augment CBF to maintain cerebral oxygen delivery, however arteriolar pathology may prevent optimal autoregulation, preventing maximal CBF. Indeed, CVR was found to be reduced in a RVCL-S cohort when cross-sectionally compared to a control cohort.⁶ Further, CBF in non-lesioned WM may be reduced in the region surrounding WMH as connecting axons degenerate. Another possible explanation is due to the challenge of accurate CBF measurements in the deep WM due to low signal-to-noise ratio giving a “floor effect” and exhibiting high intra-individual variability with repeated measurements.⁶⁰

Taken together, the findings suggest that increased OEF, as a metric of tissue hypoxia, may serve as a predictive biomarker to identify patients with early onset or an accelerated disease course in whom early or novel treatment should be considered. The finding that nOEF is associated with future WM infarction and microstructural decline supports hypoxia-ischemia as a critical, early mechanism underlying RVCL-S pathophysiology. Given the frank enhancement and pseudotumor formation commonly seen in RVCL-S, there are likely additional mechanisms such as endothelial barrier disruption, as can be measured using MRI-metrics of blood brain barrier permeability, which play a role in progression and may interact with hypoxia-ischemia to accelerate VCID.⁶¹ While validated structural neuroimaging endpoints of cSVD progression are critical as surrogate endpoints of cognitive decline (discussed below), markers of active physiology which are reversible, such as OEF, may inform pathophysiology and may be responsive to therapeutic efficacy even before structural imaging metrics demonstrate irreversible, downstream injury.

Moving towards clinical trial preparedness in RVCL-S

Recent work in cellular and animal models and in patients demonstrate that TREX1 dysfunction leads to double-stranded DNA breaks and DNA repair damage which may incite the early and accelerated systemic microvasculopathy.⁶² As we make progress towards potential therapies, preparing for clinical trials becomes increasingly important. For therapies which may be only mildly to moderately effective, discerning whether a therapy delays progression is challenging, especially if there are insufficient patients to have a contemporaneous control cohort. Thus, selecting appropriate biomarkers and clinical outcome assessments (COA) may help determine clinical efficacy in single-arm trials with small sample sizes.⁶³ Optimal COAs should translate clinically meaningful information and demonstrate high reliability and generalizability when possible.⁶⁴ For emerging, high-risk interventions, like immunosuppressive therapies, stem cell transplants, or genetic cures, patient selection and timing of initiation are crucial to prevent unnecessary adverse effects.

Given healthy young adults should have minimal WM lesions, we modelled individual WMH lesion growth trajectories to estimate an age of cerebral lesion development in RVCL-S, finding a median age of 30 years, approximately a decade earlier than estimated in previous reports.^3,5,8 Furthermore, we found substantial phenotypic variability in the appearance of WMH with an interquartile range between 24 and 46 years. Given this variability in age of lesion development, followed by uniformly, steep WMH progression in most patients, there may be an environmental or other trigger for cerebral disease onset followed by progression. To assess which neuroimaging endpoints were clinically meaningful, and thus may serve as surrogate endpoints, we examined which predicted decline in cognition. All neuroimaging endpoints were associated with progression in one cognitive measure and have been associated with cognitive decline in sporadic cSVD of aging and hypertension.^46,47 Moreover, these structural neuroimaging endpoints can be acquired on a standard 3-Tesla clinical brain MRI, thus increasing their generalizability.

Of cognitive endpoints to be considered for future clinical trials, DSST, a measure of working memory and processing speed, showed the earliest and steepest decline (−13.1 T-score points per decade), decreasing in patients as young as the early 40s (Figure 1, Table 2). Decline in DSST was predicted by increasing log-transformed normalized WMH volume and decrease in normalized WM volume. While our study population could complete the cognitive battery, future trials for this disease—where vision loss progresses to functional blindness in some cases—should include cognitive tests less dependent on vision, such as category fluency (included) and the Paced Auditory Serial Addition Test (currently not included). We found that MoCA was relatively insensitive to subtle, early changes in cognition within the RVCL-S cohort. We and others postulate that MoCA may lack sensitivity due to its absence of a processing speed subtest.^65,66

Study strengths include its prospective, longitudinal design with a uniform protocol permitting direct comparison of imaging and cognitive results across time-points and patients. MRI techniques and cognitive testing included measures specific for cSVD and VCID. Study limitations include small size, single-center approach, and use of advanced MRI methods, which limit its generalizability. International collaboration would permit capturing phenotypic variability. For example, in CADASIL, different genetic variants cause variability in disease severity and mortality.⁶⁷ Larger cohorts would permit understanding how comorbidities, environmental factors, and systemic disease, such as anemia and kidney disease, independently contribute to cerebral disease progression. Future studies should include examination of GM, cortical thickness, and smaller brain regions to understand regional heterogeneity in CBF and OEF with respect to structural changes. The control cohort is limited by: (1) recruitment of healthy individuals who volunteered rather than being randomly selected, and (2) absence of a longitudinal assessment to assess to what degree progression in RVCL-S represents accelerated vascular aging. While using the same imaging protocol to accurately quantify change in imaging endpoints over seven years is a strength, this is also a limitation, as 2D FLAIR and single-shell DTI were used without technical updates. Both 3D FLAIR and multishell DTI would improve accuracy of WMH quantification and enable measures of structural connectivity, respectively. Finally, pCASL-defined CBF is limited by low signal-to-noise ratio, especially in deep white matter which may have obscured true predictive abilities of CBF.

Conclusion

Measures of cerebral OEF, in non-lesioned WM, provide direct evidence for tissue hypoxia-ischemia as a disease mechanism in RVCL-S. Further, nOEF may provide a harbinger of disease progression and may serve as a predictive biomarker in inherited and sporadic vascular dementias. Longitudinal, observational cohort studies will continue to help establish the natural history of RVCL-S, while informing biomarker and endpoint selection for future clinical trials.