Dynamic Fusion of SNP and FNC in UK Biobank Reveals Static and Time-varying Manifolds

659 

Abstract Submission 

Jiayu Chen¹, Armin Iraji², Zening Fu³, Pablo Andrés-Camazón⁴, Bishal Thapaliya¹, Jingyu Liu¹, Vince Calhoun⁵

¹GSU, Atlanta, GA, ²Georgia State University, Atlanta, GA, ³Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS) Georgi, Atlanta, GA, ⁴Hospital General Universitario Gregorio Marañón, Madrid, Spain, ⁵GSU/GATech/Emory, Decatur, GA

Jiayu Chen  
GSU
Atlanta, GA

Armin Iraji  
Georgia State University
Atlanta, GA

Zening Fu  
Tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS) Georgi
Atlanta, GA

Pablo Andrés-Camazón  
Hospital General Universitario Gregorio Marañón
Madrid, Spain

Bishal Thapaliya  
GSU
Atlanta, GA

Jingyu Liu  
GSU
Atlanta, GA

Vince Calhoun  
GSU/GATech/Emory
Decatur, GA

Psychiatric disorders are highly heritable1 and genetic factors likely exert influence on clinical manifestations by affecting the brain2. Despite holding promise for individualized treatment, characterizing genetic and neurobiological profiles and their interrelationships has proven a big challenge for complex brain disorders, given the polygenic nature and modest effect sizes3. The current work proposes a novel dynamic fusion framework to perform multiple single nucleotide polymorphisms (SNPs) - dynamic functional network connectivity (dFNC) fusions to evaluate static and time-varying manifolds. We showcased how the proposed framework, coupled with the large UK Biobank (UKB) and aggregated schizophrenia (SZ) cohorts, offered additional insights into how genetic risk links to SZ-related dysconnectivity.

We used the QC'ed SNPs and resting fMRI (rsfMRI) data of 32,861 non-related European ancestry individuals of UKB (47% males, aged 45-81). The PGC 287 loci4 pruned at r2 < 0.2 yielded 12,946 SZ-risk SNPs. The rsfMRI data were processed using the fully automated NeuroMark pipeline5. The Neuromark_fMRI_1.0 network template served as a reference in a spatial-constrained independent component analysis6 to derive 53 intrinsic connectivity networks (ICNs), based on which windowed FNC (wFNC) was estimated7. K-means clustering on wFNCs identified four dynamic states using the elbow criterion. Next, the state-specific dFNC was computed as the mean of wFNCs assigned to that state for each subject. Joint ICA8 was then applied to the concatenated SZ-risk SNPs and each of the four state-specific dFNC matrices (32,861 × 14,324) resulting in four parallel fusions of 35 joint SNP-dFNC components.

Joint components were validated for relevance to SZ by projecting their dFNC parts to the state-specific dFNC features derived in the same way in an aggregated SZ cohort consisting of 1,237 individuals (43% SZ, 60% males, aged 16-79). A two-sample t-test identified SZ-discriminating dFNC components (FDR p < 0.05).

For each component, we evaluated its modality-specific similarity with those of other three fusions , with low similarity indicating high state-specificity, and hence high dynamism. For validated SZ-relevant components, we identified the top SNPs and dFNC features (|z-score|>3). Gene Ontology pathway analysis was conducted on the annotated genes of top SNPs for enriched biological processes. The top dFNC features (i.e., connectivity edges) were interpreted based on the anatomical labels of the involved ICNs.

A wide range of dynamism was noted for both SNP and dFNC modality across four parallel fusions (Fig. 1a-1b). A total of 53 joint components were validated as SZ-relevant, which did not appear to be biased towards high or low dynamism. Fig. 1c-1f present component 9 of SNP-dFNCstate1 fusion (State1_jICA9) and State3_jICA7 as an example of intermediate similarity. Both dFNC components highlighted thalamus-seeded connections, and their SNP components shared 37% of the annotated genes, both enriched for cell projection organization. In contrast, Fig. 1g-1h shows State1_jICA4 as an example of low similarity. Its SNP component was enriched for cell projection also, despite not being correlated with State1_jICA9. Its dFNC component highlighted insula-seeded connections.

Thalamus and insula dysconnectivity are well documented in SZ9,10. Cell projection organization results in the arrangement of constituent parts, disassembly of a prolongation, or process extending from a cell, e.g., a flagellum or axon. Notably, the State1_jICA4 SNPs, which might be elicited only upon dynamic fusion with dFNC given its high dynamism, might be complementary to State1_jICA9 to provide a more complete picture of the genomic factor conferring risk to SZ by affecting cell projection. This presents the benefit of the proposed dynamic fusion to expand the SNP data for time-varying manifolds that may provide additional insights into the underlying biology.

Psychiatric (eg. Depression, Anxiety, Schizophrenia) ¹

Genetic Modeling and Analysis Methods ²

fMRI Connectivity and Network Modeling

Multivariate

Psychiatric

Schizophrenia

Thalamus

Other - fusion