Abstract This application will combine the strengths of two large scale NIH-funded initiatives to understand disorder- related patterns in the human brain: Connectomes Related to Human Disease (CRHD) and Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA). We will develop and evaluate novel brain vulnerability metrics - based on the idea of polygenic risk scores – that we expect to better predict diagnosis and cognitive performance than standard neuroimaging measures. We define a metric of “vulnerability” by quantifying the similarity between each individual's brain pattern and deficit patterns in neuropsychiatric disorders. The Regional Vulnerability Index (RVI) uses Big Data meta-analyses to quantify the similarity between an individual and meta-analytical deficit effect size patterns based on large and diverse international samples. The Machine Learning-Vulnerability Index (MVI) is trained using Big Data mega-analytic samples to quantify the similarity for individual brain patterns to those learned from patients and controls. We will compute novel, cross-domain vulnerability metrics to phenotype each of the N=3,350 CHRD individuals across three mainly psychiatric (schizophrenia-spectrum and psychosis disorder, major depression, and bipolar disorder), three mainly neurological (epilepsy, mild cognitive impairment, and Alzheimer's disease) and three neuroimaging domains (structural, diffusion, and resting state functional MRI). Our Specific Aims merge CRHD and ENIGMA data to test four hypotheses: 1) Neuropsychiatric illnesses not only impact an isolated region or circuit, but are associated with deficit patterns across multiple brain regions and circuits that can be unique to each illness; 2) such deficit patterns are informative of cognitive deficits in patients; 3) similarity with the deficit patterns will have higher specificity and sensitivity than any traditional neuroimaging metric or trait; and 4) similarity at the voxel- and vertex-based level may lead to development of high-resolution vulnerability indices. We will test these hypotheses by performing patient-control sensitivity and specificity analyses in the corresponding CRHD illness groups; study the degree of separation vs. commonality across psychiatric and neurological disorders; and evaluate pattern differences at specific stages of the illnesses, such as Alzheimer's disease. We will use multivariate mediation analyses to link vulnerability to variance in cognitive domains ascertained by CHRD. This short, intensive project will benefit the community-at-large by populating the CHRD/HCP database with novel multimodal brain phenotypes extracted and homogenized using standard ENIGMA workflows, enriched with Open Science approaches.

PROJECT NARRATIVE/RELEVANCE The Connectomes Related to Human Disease/Human Connectome Project (CRHD) and Enhancing Neuroimaging Genetics through Meta-Analysis (ENIGMA) are two large scale NIH-funded initiatives exploring disorder-related patterns in the human brain. The two projects are complementary: CRHD focuses on in-depth, high-resolution multimodal phenotyping of the brain connectome in modest samples, whereas ENIGMA's analyses harness the statistical power of large and inclusive, worldwide samples. We propose combining the strengths of CRHD and ENIGMA to test novel biomarkers based on replicable and consistent illness deficit patterns provided by Big Data studies. We posit that major brain illnesses do not impact an isolated region or brain circuit but rather lead to deficit patterns across multiple brain regions and circuits that are shared or unique to an illness; such deficit patterns are highly replicable and linked to severity, duration and treatment resistance. We will develop and evaluate a multivariate vulnerability model based on (1) the degree of similarity between an individual's brain patterns and expected disorder patterns defined as a linear combination of deficits - conceptually similar to polygenic risk scores - and (2) a vulnerability metric computed using machine learning. We will test the specificity and sensitivity of these models against the single most discriminative neuroimaging traits in six diagnostic categories shared by CRHD and ENIGMA: schizophrenia-spectrum disorders (SSD) and psychosis, major depressive disorder (MDD), bipolar disorders (BP), epilepsy (EP), mild cognitive impairment (MCI), and Alzheimer's disease (AD). We will use multivariate analyses to compare sensitivity and specificity of the regional vulnerability index (RVI) and machine learning vulnerability (MVI) in separating patients and controls in the CHRD samples, mapping unique and shared deficit patterns across six diagnostic categories. We will link vulnerability metrics to cognitive outcomes and diagnosis using mediation analyses. Finally, we will enrich vulnerability models with CHRD's connectomics based phenotypes including cortical myelination to advance white matter integrity measurements for future large-scale ENIGMA studies.