ABSTRACT
White matter hyperintensities (WMH) are an important factor in the occurrence and progression of stroke,
cognitive decline, and dementia in the aging population, but pose a puzzling challenge in evaluations of
brain health and disease due to its underlying pathologic heterogeneity. While WMH are frequently
considered as a consequence of cerebral small-vessel disease (CSVD), emerging evidence suggest WMH
may also arise from non-vascular processes including Alzheimer’s disease (AD)-related neurodegeneration
or neuroinflammation. Understanding the multifactorial etiologies of WMH is critical for development of much
needed therapies and preventative strategies, especially given its high prevalence in older community
persons. However, commonly used assessment methods based on global severity burden fail to address
such heterogeneity. WMH spatial patterns is a novel phenotype that can be extracted from structural MRI
data using advanced pattern analysis methods that capitalizes on variability in WMH topography across
different diseases. We previously show that machine learning (ML)-derived spatial patterns of WMH more
precisely capture the underlying heterogeneity in WMH pathology compared to standard region or lobar-
based classifications. Distinct ML-derived WMH spatial patterns have unique associations with different
WMH etiologies, distinguishing between WMH arising from CSVD subtypes (cerebral amyloid angiopathy
vs. hypertensive), AD, and normal aging. We hypothesize that focusing on ML-derived WMH spatial patterns
will greatly expand the discovery of genetic contributions to WMH, leading to better understanding of the
molecular basis of WMH and development of mechanism-based treatments. To this end, we will leverage
existing neuroimaging and genetic data resources from 7 cohorts enriched for diverse conditions with high
WMH prevalence (n=4,872) as the discovery cohort, and the UK Biobank (UKB, n>60,000) representing the
general population. This will allow us to derive distinct, disease-specific WMH spatial patterns from
neuroimaging data using machine learning, examine their biological correlation with vascular and amyloid-
related traits, and evaluate relevance of ML-derived WMH spatial patterns for predicting long-term
development of stroke or dementia (Aim 1). We will integrate molecular quantitative trait loci (QTL) data into
genome-wide association analysis (GWAS) of ML-derived WMH spatial patterns to prioritize identification of
causal, functionally active genes (Aim 2). We will perform genomics-driven drug discovery using Mendelian
randomization (MR) to identify therapeutic targets relevant to WMH (Aim 3). This project leverages the
combined expertise of the PI and assembled team in neuroimaging, genomics, and informatics. Completion
of project aims will greatly advance our understanding of genetic contributions to WMH, providing novel
mechanistic insights into the role of white matter lesions in brain health, accelerate therapeutic discovery by
identification of drug repositioning opportunities, and lay the groundwork for personalized diagnosis and
care by disentangling the heterogenous nature of WMH at the individual level.
