Association Between Vascular NOTCH3 Aggregation and Disease Severity in a CADASIL Cohort - Implications for NOTCH3 Variant-Specific Disease Prediction

Abstract

Objective: Vascular NOTCH3 protein ectodomain aggregation is a pathological hallmark of cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL), a monogenic small vessel disease typically caused by cysteine-altering variants in NOTCH3. Given their high population frequency, these NOTCH3 variants are an important genetic contributor to stroke and vascular dementia worldwide. Disease severity in CADASIL is highly variable and is mainly determined by the position of the pathogenic NOTCH3 variant in the NOTCH3 ectodomain. Here, we aimed to investigate the association between NOTCH3 aggregation load in skin vessels, cysteine-altering NOTCH3 variants, and disease severity in a prospective cohort study of 212 patients with CADASIL with 39 distinct cysteine-altering NOTCH3 variants.

Methods: NOTCH3 aggregation load in skin vessels was determined by calculating the NOTCH3 score; the fraction of skin vessel wall area positive for NOTCH3 staining. Variant-specific NOTCH3 scores were calculated for variants present in 10 or more participants, by averaging the NOTCH3 scores of individuals with that distinct variant. The associations between the NOTCH3 score, NOTCH3 variants, and neuroimaging and clinical outcomes were investigated using multivariable linear mixed models, Cox regression, and mediation analyses.

Results: The NOTCH3 score was significantly associated with lifetime stroke probability and small vessel disease neuroimaging outcomes, but not with age. Variant-specific NOTCH3 scores reflected differences in disease severity associated with distinct NOTCH3 variants.

Interpretation: These findings suggest that differences in NOTCH3 aggregation propensity underlie the differences in disease severity associated with NOTCH3 cysteine-altering variants, and show that NOTCH3-variant specific NOTCH3 scores can contribute to improved individualized disease prediction in CADASIL. ANN NEUROL 2025;98:273-285.

FIGURE 1

The association between the NOTCH3 score in skin vasculature with clinical and neuroimaging outcomes and with age. (A) Examples of NOTCH3 score measurements in two blood vessels. (1) Schematic overview of a skin biopsy, with the dermal and subcutaneous layers in gray and the epidermal layer in green. (2) Original images of blood vessels stained with a NOTCH3^ECD‐antibody (upper panel scale bar = 10 μm and lower panel scale bar = 20 μm). (3) Region of interest drawn in pink (the blood vessel wall). (4) Thresholding of NOTCH3^ECD positive area. The NOTCH3 score was calculated by dividing the NOTCH3^ECD‐positive area by the total vessel area, and then taking the average for the 10 most highest scoring vessels per sample. (B–G) Scatterplots with the standardized NOTCH3 score on the x‐axis (after square root transformation) and standardized neuroimaging outcomes on the y‐axis (conditional Pearson's residuals derived from linear mixed effects models corrected for age, sex, cardiovascular risk factors, and batch effects). There was a significant correlation between the NOTCH3 score and PSMD (p = 6.0 × 10⁻¹¹, p _adj = 4.2 × 10⁻¹⁰ [B]), nWMHv (p = 5.4 × 10⁻¹², p _adj = 4.3 × 10⁻¹¹ [C]), nLV (p = 1.5 × 10⁻⁷, p _adj = 8.8 × 10⁻⁷ [D]), CMB count (p = 0.0017, p _adj = 0.0057 [E]), and PVS score (p = 0.0014, p _adj = 0.0057 [F]) but not with BPF (p = 0.72, p _adj = 1 [G]) after correction for age, sex, and cardiovascular risk factors and batch effects. (H) Kaplan–Meier plot showing the differences in stroke probability between the tertiles of the NOTCH3 score (with the first tertile containing those with the lowest score, and the third tertile the highest). A higher NOTCH3 score was associated with a higher lifetime stroke probability (p = 3.1 × 10⁻⁵, p _adj = 1.5 × 10⁻⁴ [H]), after correction for sex and cardiovascular risk factors and batch effects. There was no correlation between the NOTCH3 score and TMT_B/A (p = 0.74, p _adj = 1 [I]). (J–K) There was no association between age (range = 26–81 years) and the NOTCH3 score (p = 0.75, p _adj = 1 [J]). The NOTCH3 score of brain vessels was not correlated with the age of death (p = 0.54 [K]). PSMD = peak width of the skeletonized mean diffusivity; nWMHv = normalized white matter hyperintensity volume; nLV = normalized lacune volume; CMB = cerebral microbleed; PVS = perivascular space; BPF = brain parenchymal fraction; NOTCH3^ECD = NOTCH3 ectodomain; TMT_B/A = Trail Making Test B Given A t scores; p _adj = multiple correction adjusted p value (using the Holm method, n = 8 tested hypotheses for neuroimaging and clinical outcomes; n = 7 tested hypotheses for modifiers of the NOTCH3 score). Levels of significance: **** = p < 0.0001, *** = p < 0.001, ** = p < 0.01, * = p < 0.05, ns = not significant.

FIGURE 2

The association between the NOTCH3 score and NOTCH3 variant risk categories. (A) Violin plots with NOTCH3 variant risk group on the x‐axis and standardized skin NOTCH3 score per participant on the y‐axis. The standardized NOTCH3 score is equal to the conditional Pearson's residuals derived from linear mixed effects models corrected for sex, age, and CVRF and batch effects. There was a significant difference in the NOTCH3 score between participants with HR‐ and MR‐NOTCH3 variants in the main dataset (p = 5.7 × 10⁻⁹, p _adj = 4.0 × 10⁻⁸) and in the replication dataset (p = 1.5 × 10⁻⁶). (B) In brain vessels, the NOTCH3 score in the HR‐NOTCH3 group was significantly higher than in the MR‐NOTCH3 group (p = 0.0091). (C–G) The association between the NOTCH3 score and PSMD (β = 0.22, 95% CI = 0.12 to 0.31, p = 1.6 × 10⁻⁵, p _adj = 9.5 × 10⁻⁵ [C]), nWMHv (β = 0.22, 95% CI = 0.12 to 0.32, p = 1.8 × 10⁻⁵, p _adj = 9.5 × 10⁻⁵ [D]), nLV (β = 0.24, 95% CI = 0.11 to 0.37, p = 6.0 × 10⁻⁴, p _adj = 0.0024 [E]), and CMB count (β = 0.17, 95% CI = 0.039 to 0.29, p = 0.013, p _adj = 0.039 [F]) remained significant after stratification for NOTCH3 variant risk category, but not PVS score (β = 0.13, 95% CI = −0.015 to 0.28, p = 0.087, p _adj = 0.087 [G]). PSMD = peak width of the skeletonized mean diffusivity; nWMHv = normalized white matter hyperintensity volume; nLV = normalized lacune volume; CMB = cerebral microbleed; PVS = perivascular space; HR‐NOTCH3 = high‐risk NOTCH3 variant risk category; MR‐NOTCH3 = moderate‐risk NOTCH3 variant risk category; LR‐NOTCH3 = low‐risk NOTCH3 variant risk category; p _adj = multiple correction adjusted p value (using the Holm method, n = 6 tested hypotheses for neuroimaging and clinical outcomes; n = 7 tested hypotheses for modifiers of the NOTCH3 score). Levels of significance: **** = p < 0.0001, *** = p < 0.001, ** = p < 0.01, * = p < 0.05, ns = not significant.

FIGURE 3

The association between the NOTCH3 score and distinct NOTCH3 variants. (A) A schematic overview of the NOTCH3 protein. The ECD of the NOTCH3 protein includes 34 EGFr domains. Each EGFr domain contains numerous distinct NOTCH3 variants that may lead to CADASIL, the majority of which are cysteine‐altering missense variants. Distinct NOTCH3 variants which were represented by 10 or more participants are denoted by symbols above the EGFr domains in which they are located. NOTCH3 variants are classified into NOTCH3 variant risk categories based on the EGFr domain in which they are situated, and can be classified into high risk (HR‐NOTCH3, in red), moderate risk (MR‐NOTCH3, in blue), low risk (LR‐NOTCH3, in green) or unknown (UR‐NOTCH3, in gray) categories. These NOTCH3 variant risk categories were published elsewere and are based on cohort‐to‐population frequency odds ratios of the EGFr domains these NOTCH3 variants are located in. (B) Violin plots of the NOTCH3 score with on the x‐axis distinct NOTCH3 variants grouped per EGFr domain (in red HR‐NOTCH3 variants; in blue MR‐NOTCH3 variants; and in green, LR‐NOTCH3 variants) and on the y‐axis, the standardized NOTCH3 score (conditional Pearson's residuals derived from mixed effects models corrected for age, sex, CVRF, and batch effects). The black dots represent the NOTCH3 score for one individual participant; the orange dot represents the average for a single NOTCH3 variant. The red line represents the average for all HR‐NOTCH3 variants combined, whereas the blue line is the average for all MR‐NOTCH3 variants combined. Differences between EGFr domains accounted for a significant proportion of variability observed in the NOTCH3 score in the HR‐ and MR‐NOTCH3 groups (p = 1.0 × 10⁻¹⁹). Some of the NOTCH3 variants had a striking deviation of the NOTCH3 score from their respective NOTCH3 variant risk categories, including the HR‐NOTCH3 variants p.(Arg207Cys; n = 33), p.(Arg449Cys; n = 2), and the MR‐NOTCH3 variants in EGFr domains 9 and 10 (n = 7), and p.(Arg1076Cys; n = 4). As some variants are overrepresented in EGFr domains, and due to power considerations, further analyses of disease severity were performed using only those distinct NOTCH3 variants with 10 or more participants (C). (C) The variant‐specific NOTCH3 score (variant NOTCH3 score) with 95% confidence intervals, sorted in ascending order. The blue and the red lines indicate the NOTCH3 variant risk category (blue = MR‐NOTCH3; red = HR‐NOTCH3). (D) Images depicting blood vessels with representative granular NOTCH3^ECD staining with a NOTCH3 score equal to that of their respective variant. The images of p.(Arg578Cys) and p.(Arg207Cys) variants show similar NOTCH3^ECD aggregation, whereas the p.(Arg141Cys) variant shows more NOTCH3^ECD aggregation. EGFr = epidermal growth factor‐like repeat; HR‐NOTCH3 = high‐risk NOTCH3 variant risk category; MR‐NOTCH3 = moderate‐risk NOTCH3 variant risk category; LR‐NOTCH3 = low‐risk NOTCH3 variant risk category; UR‐NOTCH3 = unknown‐risk NOTCH3 variant risk category. ^aThe p.(Arg544Cys) variant is located between EGFr domains 13 and 14.

FIGURE 4

The association between variant‐specific NOTCH3 scores and disease severity. (A, B) Upper panels. The variant‐specific NOTCH3 score (variant NOTCH3 score) was significantly associated with PSMD (p = 6.3 × 10⁻¹³ [D]), nWMHv (p = 1.4 × 10⁻¹⁷ [E]), nLV (p = 4.2 × 10⁻¹⁴ [F]) and performed better than using an individual participant's NOTCH3 score in predicting disease severity (see Supplementary Table S4). (A, B) Lower panels. EMMs with 95% confidence intervals of PSMD (D), nWMHv (E), and nLV (F) per variant, arranged in ascending order of the variant‐specific NOTCH3 score. The blue and red lines underneath each datapoint represent the NOTCH3 variant risk category (red = HR‐NOTCH3, blue = MR‐NOTCH3). EMM = estimated marginal means; PSMD = peak width of the skeletonized mean diffusivity; nWMHv = normalized white matter hyperintensity volume; nLV = normalized lacune volume. Levels of significance: **** = p < 0.0001.

FIGURE 5

The NOTCH3 score and disease severity of the p.(Arg207Cys) variant. Using a variant‐specific NOTCH3 score‐first approach may provide a framework for increasing the resolution of clinically relevant risk classification on the level of distinct NOTCH3 variants. The HR‐NOTCH3 p.(Arg207Cys) variant was associated with a lower variant‐specific NOTCH3 score than the average of other HR‐NOTCH3 variants (p = 4.9 × 10⁻¹¹), and did not differ from the average NOTCH3 score of the MR‐NOTCH3 group (p = 0.60 [A]). The p.(Arg207Cys) variant was significantly less severe than the average of the HR‐NOTCH3 group for PSMD (p = 1.6 × 10⁻³ [B]), nWMHv (p = 4.8 × 10⁻⁶ [C]), nLV (p = 1.1 × 10⁻⁶ [D]), and lifetime stroke probability (p = 0.0061 [E]), and more severe than the MR‐NOTCH3 group for PSMD (p = 7.2 × 10⁻⁵ [B]), nWMHv (p = 2.2 × 10⁻⁵ [C]), and lifetime stroke probability (p = 0.045 [E]). PSMD = peak width of the skeletonized mean diffusivity; nWMHv = normalized white matter hyperintensity volume; nLV = normalized lacune volume; HR‐NOTCH3 = high‐risk NOTCH3 variants; MR‐NOTCH3 = moderate‐risk NOTCH3 variants; Levels of significance: **** = p < 0.0001, ** = p < 0.01, * = p < 0.05, ns = not significant.