Association of NOTCH3 Variant Position With Stroke Onset and Other Clinical Features Among Patients With CADASIL

Longitudinal data were collected from patients with CADASIL over a 30-year period, as part of a UK CADASIL National Referral Service from 1990 to 2015 and through enrolment to the UK Familial Cerebral Small Vessel Disease Study from 2016 to 2021. Data from both sources were collected through a uniform approach on a standard proforma.

485 patients were included in this study. They were evaluated and examined by consultant neurologists as part of their clinical appointment. The collected data were reviewed by a neurologist, and major disease events were confirmed by referring to original hospital records.

All the patients had confirmed genetic diagnosis through direct sequencing of the NOTCH3 gene to identify the pathogenic variants for CADASIL, namely missense variants, leading to the gain or loss of a cysteine residue in 1 of the 34 EGFR domains of the NOTCH3 protein. Patients who were diagnosed with CADASIL by presymptomatic genetic testing were also included (n = 22). Two patients with confirmed homozygous or compound heterozygous variants on the NOTCH3 gene were excluded because they may have different clinical features that mask the effects of variant position on disease severity.⁹

The following fields were collected for all patients: demographics (year of birth and sex), NOTCH3 variant information, first presenting feature and age at which it was experienced, history of and age at onset of various diseases and symptoms (such as stroke and encephalopathy), and history of cardiovascular risk factors (namely hypertension and smoking). Brain metrics, including total brain volume, WMH volume, lacune count, and microbleed count were available if the patient had a clinical MRI. Brain and WMH volumes were recorded in cm³.

Stroke was diagnosed based on clinical and neuroradiologic assessment and defined as a focal neurologic event lasting more than 24 hours accompanied by brain imaging evidence. Migraine was defined by the 2004 International Classification of Headache Disorders.¹⁰ Epileptic seizures were only included if they occurred spontaneously and not as part of an encephalopathic episode. History of any psychiatric illness, hypertension, hypercholesterolemia, or diabetes was recorded. CADASIL encephalopathy was defined as in a previous article as an acute reversible encephalopathy with evidence of reduced consciousness in the absence of any other organic cause, where symptoms lasted for longer than 24 hours and were sufficient to warrant hospital admission.¹¹ History of depression, defined as depression requiring treatment with drug or psychological therapy, and symptomatic cognitive impairment was determined from self-reported records taken during clinical assessment. Hypertension was defined as a systolic blood pressure above 140 mm Hg and/or a diastolic pressure above 90 mm Hg or use of antihypertensive medication. Hypercholesterolemia was defined as a serum cholesterol greater than 5.2 mmol/L or on lipid-lowering drugs.

All participants in this study gave written informed consent. The UK Cerebral Small Vessel Disease Study was approved by the East of England–Cambridge Central Research Ethics Committee (Ref: 16/EE/0118).

Brain MRI scans acquired during routine clinical management were available for some patients. Some of the scans were excluded from further analysis for each of the MRI parameters studied (WMH volume, lacune, and microbleed quantification) because of quality issues and/or unavailable MRI sequence. MRI was performed at the patient's local hospital at field strengths of between 0.5 and 3 T. T1-weighted and fluid-attenuated inversion recovery (FLAIR) T2-weighted sequences were used to calculate brain and WMH volumes, respectively.

Brain volumes were estimated using SIENAX¹² from the FMRIB's software library 5.0.9.¹³ Brain tissue was extracted using an automated Brain Extraction Tool¹⁴ and linearly registered using affine transformation,^15,16 which helps to normalize volumes for skull size. Tissue-type segmentation of images was performed through a hidden Markov random field model and the expectation-maximization algorithm.¹⁷ Normalized measures of gray matter, white matter, and total brain volumes were calculated by multiplying the measured volume by its estimated scaling factor.

Quantification of WMH on FLAIR images was performed using a semiautomated contouring method in the JIM software package version 8.0 (Xinapse Systems, Colchester, United Kingdom). WMH volumes were marked and measured on T2 FLAIR sequences in the axial plane by a trained researcher. Final volumes were multiplied by the scaling factor derived from the brain volume to account for skull size. The WMH lesion volumes were marked on 10 randomly selected MRI scans by 2 independent raters. The intraclass correlation (ICC) between the 2 raters was 0.96.

Lacunes and microbleeds were quantified by a neurologist. Lacunes were manually counted on FLAIR images and distinguished from perivascular spaces based on their location, size, and shape on T1-weighted and T2-weighted scans. Lacunes were counted if they were in the subcortex and larger than 3 mm in diameter. Microbleeds were counted using the Brain Observer Microbleed Scale, defined as round well-defined hypointense foci on susceptibility weighted imaging or gradient echo sequences with a diameter of 2–10 mm.¹⁸ The inter-rater agreement for lacune and microbleed counts were good, ICC for count data = 0.91 and 0.99, respectively.

Several bioinformatic tools were used to predict the molecular effects of NOTCH3 variants at different positions. Variants were annotated using Ensembl Variant Effect Predictor.¹⁹ They were mapped onto protein 3-dimensional structures in the Protein Data Bank on VarSite,²⁰ whose information was analyzed to predict the molecular effects of the variants. Protein aggregation scores were calculated based on a small part of the NOTCH3 protein structure using the Aggrescan3D web server version 2.0,²¹ with the distance of aggregation analysis specified at 5 Å. The intrinsic aggregation propensity of each amino acid was computed and used to generate the average aggregation score of the protein structure. The difference in residue aggregation scores between the mutated NOTCH3 protein and the wild-type protein was calculated for each NOTCH3 variant. Proaggregation variants were defined as variants causing an increased residue score (>0).

Baseline characteristics and cardiovascular risk factors were compared between patients with variants in EGFRs 1–6 and 7–34 by 2-sample t test. Any variables that did not follow a normal distribution were transformed: Brain and WMH volumes were natural-log transformed; lacune and microbleed counts were square rooted. The association of EGFR group with risk of various clinical features was analyzed by logistic regression with adjustment for age, sex, hypertension, and smoking, and the effects of variant position on MRI measures were analyzed by linear regression with adjustment for age, sex, and MRI field strength. Kaplan-Meier analyses and Cox proportional hazards regression (using age as the underlying time scale) with adjustment for sex and cardiovascular risk factors were performed to compare the time of onset of disease in different EGFR groups. These analyses incorporated all patients and censored those who had not experienced an event at the time of their most recent clinical assessment.

Our initial a priori analysis was based on previous literature suggesting EGFRs 1–6 variants had a more severe phenotype than EGFRs 7–34 variants.⁷ After this comparison, further exploratory analyses were performed with NOTCH3 variants that were further divided based on groupings other than EGFRs 1–6 and 7–34 to investigate whether any other EGFR cutoff better separated patients into high-risk and low-risk groups for stroke. Variants in EGFRs 1–6 were also subdivided into 3 smaller groups (EGFRs 1–2, 3–4, and 5–6) to determine whether associations were stronger within any subdivision of this EGFR category. The hazard ratios (HRs) of stroke for these groups were calculated using the Cox proportional hazards model, where assumptions were tested based on Schoenfeld residuals, and no violations were observed. We used floating variances to estimate 95% CIs for each group (including the reference group) that reflect the amount of information underlying each category.²²

The association of gain-of-cysteine and proaggregation variants on stroke outcomes was investigated. Both types of variants were correlated with EGFR position (1–6 vs 7–34) using logistic regression (adjusted for age and sex) and with the age at onset of the first stroke using Cox regression (adjusted for sex and EGFRs 1–6 variant). To assess whether gain-of-cysteine and proaggregation variants mediated the effect of EGFR position on stroke, mediation analyses were performed using the “mediation” package in R.²³ Standard errors of the indirect effect (i.e., the association of variant on risk of stroke acting through the molecular mediator considered) were obtained using 10,000 bootstrap samples.

All statistical analyses were performed using R V.4.0.3 with 2-sided p values, and p < 0.05 considered statistically significant.

The patient data used for this study are available from the corresponding author on reasonable request.

A total of 485 patients from 345 different families with a genetically confirmed CADASIL diagnosis were included in this study. All patients carried heterozygous NOTCH3 variants. The demographic and risk factor profile of patients are summarized in eTable 1 (links.lww.com/WNL/C74).

There were 76 different NOTCH3 variants spread across 16 different exons (between exons 2 and 24 of the NOTCH3 gene), with most in exons 3 and 4 (n = 358, 76.8%) (Figure 1 and eTable 2, links.lww.com/WNL/C75). Most of the variants were in the proximal region of NOTCH3 encoding EGFRs 1–6 (n = 391, 80.6%, see eFigure 1, links.lww.com/WNL/C71). Variants in EGFRs 3 and 4 comprised 28.7% (n = 139) and 32.4% (n = 157) of the cohort, respectively. Thirty-two of 76 (42.1%) variants led to the gain of cysteine residue in an EGFR domain, whereas 44 (57.9%) led to the loss of a cysteine (Figure 1 and eTable 2).

Patients with an EGFRs 1–6 variant were on average 5.1 years younger than patients with an EGFRs 7–34 variant (49.1 years vs 54.2 years, p = 0.0002) (eTable 1, links.lww.com/WNL/C74). After adjusting for age and sex, variants in EGFRs 1–6 were associated with a more than twofold increase in the odds of stroke (odds ratio [OR]: 2.09, 95% CI: 1.27–3.49, p = 0.003) (Table 1).

Median age at onset of the first stroke was 55 (interquartile range [IQR]: 13) years for EGFRs 1–6 patients and 64 (IQR: 18) for EGFRs 7–34 patients. Using Cox regression after adjustment for sex and cardiovascular risk factors, variant location remained a strong predictor of age at stroke onset (HR for EGFRs 1–6 variant vs EGFRs 7–34 variant: 2.05, 95% CI: 1.43–2.94, p = 8.5 × 10⁻⁵) (Figure 2). Family of origin did not alter the effect of NOTCH3 variant position on age at stroke onset (i.e., with the inclusion of only 1 patient per family in our analysis, similar results were found).

Encephalopathy was more prevalent in the patients with an EGFRs 1–6 variant than those with an EGFRs 7–34 variant (OR: 2.45, 95% CI: 1.05–6.93, p = 0.04) (Table 1). In Cox models, after adjustment for sex and cardiovascular risk factors, EGFRs 1–6 variants were associated with age at first encephalopathy (HR: 2.70, 95% CI: 1.15–6.37, p = 0.02) (Figure 3).

MRI scans from 206 patients with an EGFR 1–6 variant and 55 with an EGFR 7–34 variant were visually rated and compared. For each MRI marker, the numbers of analyzable scans were brain volume (205), WMH (206), lacunes (199), and microbleeds (129). At the time of the MRI scans, EGFR 1–6 patients were on average 6.6 years younger than EGFR 7–34 patients (48.4 vs 55.0 years, p = 0.00008). In unadjusted analyses, there was a borderline association between NOTCH3 variant position and WMH (R = −0.12, p = 0.05) (eFigure 2, links.lww.com/WNL/C72); however, no differences were found when comparing the MRI markers between the 2 groups of patients (Table 2). The results were similar before and after adjustment for MRI field strength, which only correlated with microbleeds (β: 0.18, 95% CI: 0.03–0.33, p = 0.02).

The effects of different EGFRs cutoffs other than 1–6 and 7–34 on the presence of stroke were examined. Figure 4 shows the association of variants in different EGFR regions with risk of stroke. This illustrates that there was no clear difference in strength of association with stroke for EGFRs 1–9 and 18–34 but that EGFRs 10–17 seemed to be associated with a lower stroke risk compared with domains 3–4 (Figure 4). When using EGFRs 3–4 as the reference, the risk of stroke associated with EGFRs 12–17 was almost 80% lower (HR: 0.23, 95% CI: 0.12–0.47, p = 7.7 × 10⁻⁵).

Variants in the ligand-binding site (EGFRs 10–11) were also associated with a lower risk of stroke compared with variants in EGFRs 3–4 (HR: 0.34, 95% CI: 0.15–0.76, p = 0.01). MRI markers did not differ between patients with an EGFRs 10–11 variant and those with a variant outside EGFRs 10–11 (eTable 3, links.lww.com/WNL/C76).

Variants in EGFRs 1–6 were more likely to be gain-of-cysteine than loss-of-cysteine (OR: 2.73, 95% CI: 1.62–4.56, p = 0.0002). The mean age at onset of the first stroke for the gain-of-cysteine variant group did not differ from the others (restricted mean (median), 57.7 years (57 years) vs 58.8 years (59 years), p = 0.40). After adjustment for sex, no association was found between gain-of-cysteine variants and stroke onset (HR: 1.14, 95% CI: 0.82–1.60, p = 0.44). The HR remained nonsignificant after further adjusting for family of origin, cardiovascular risk factors, and EGFR group (HR: 1.07, 95% CI: 0.72–1.60, p = 0.74). Consistent with this, formal mediation analysis indicated that gain-of-cysteine variant did not mediate the effect of variant site on stroke risk (indirect effect: −0.002, 95% CI: −0.02 to 0.02, p = 0.79) (Figure 5A) or on age at stroke onset when only focusing on those who experienced a stroke (indirect effect: 0.19, 95% CI: −0.18 to 0.53, p = 0.51).

Earlier variant position was associated with a larger difference in residue aggregation scores between the mutated and wild-type protein. (eFigure 3, links.lww.com/WNL/C73). This is coherent with analyses which revealed EGFRs 1–6 variants were more likely to be proaggregation, that is, the variant was associated with a higher residue aggregation score than EGFRs 7–34 (OR: 6.13, 95% CI: 3.70–10.19, p = 3.2 × 10⁻¹²).

When focusing on aggregation status of variants, the mean age at onset of the first stroke for the proaggregation variant group (change in residue aggregation scores between the mutated and wild-type protein >0) was lower (restricted mean (median), 56.8 years (56 years) vs 62.7 years (60 years), p = 0.02). After adjustment for sex, proaggregation variants were associated with age at stroke onset (HR: 1.50, 95% CI: 1.05–2.14, p = 0.03). The HR remained significant after correction for family of origin and cardiovascular risk factors (HR: 1.63, 95% CI: 1.08–2.47, p = 0.003), but it became nonsignificant after controlling for EGFRs 1–6 (HR: 1.17, 95% CI: 0.80–1.71, p = 0.43). Furthermore, mediation analysis showed that proaggregation variant status did not mediate the effect of variant site on stroke risk (indirect effect: 0.02, 95% CI: −0.02 to 0.06, p = 0.31) (Figure 5B) or on age at stroke onset when only focusing on those who experienced a stroke (indirect effect: 0.39, 95% CI: −0.17 to 1.10, p = 0.17). Gain-of-cysteine variant did not mediate the effect of variant site on proaggregation variant status (indirect effect: 0.06, 95% CI: −0.05 to 0.16, p = 0.36) (Figure 5C). We also performed the analysis using the change in residue aggregation scores between the mutated and wild-type protein as a continuous variable instead of using the aggregation status of variants, and similarly, no evidence of mediation was found.