A73-07 Disentangling Synergistic and Independent Gene Expression Trajectories Linked to Emphysema and Small-airway Function in COPDgene

Abstract Rationale Chronic Obstructive Pulmonary Disease (COPD) is characterized by substantially heterogeneous molecular and physiological trajectories. Traditional gene-level analyses often overlook higher level mechanisms of molecular programs that act jointly or independently over time. To better characterize the complex dynamics of molecular mechanisms in COPD, we employ a latent factor graph framework for longitudinal data to disentangle independent and synergistic components of gene expression (GE) variation, thereby revealing the GE trajectories associated with lung outcomes. Methods Blood-derived RNA-seq data (14,892 genes) from 3,982 Phase 2 and 510 Phase 3 COPDGene participants were used in the study. We extracted latent gene programs that summarize GE patterns using the LOVE algorithm, which allows features (genes) to be members of more than one latent factor. Two model assumptions were evaluated: (1) independent, and (2) synergistic latent factor interactions. The former assumes that each factor contributes independently, whereas the latter allows for interactions. Next, we used a new algorithm we developed (Yuan et al, in preparation) to calculate the trajectory statistic for the GE and clinical variables. The new transform algorithm takes into account asynchronicity and undersampling of measurements. Causal graph learning algorithm was used for modeling potential cause-effect interactions between gene programs and clinical variables. For significantly enriched pathway analysis we used Enrichr. Results In the independent model, upper lobe percent emphysema was found to be directly connected to latent factor V170 (Fig. 1A), whose genes are enriched for cell types such as type I pneumocytes, neutrophils, and lymphoid stem cells. In the synergistic model, V305 is directly linked to predicted FEF25-75%. V305 latent factor genes are enriched in pathways related to nicotine degradation, Wnt signaling network, and cadherin signaling pathway. Conclusions By contrasting synergistic and independent latent factor networks, this study demonstrates how incorporating synergistic effects in longitudinal GE modeling uncovers varying coordinated molecular processes underlying lung outcomes. Funding This work was supported by NHLBI R01HL157879. The COPDGene study (NCT00608764) is supported by grants from the NHLBI (U01HL089897 and U01HL089856), by NIH contract 75N92023D00011, and by the COPD Foundation through contributions made to an Industry Advisory Committee that has included AstraZeneca, Bayer Pharmaceuticals, Boehringer-Ingelheim, Genentech, GlaxoSmithKline, Novartis, Pfizer and Sunovion. Figure 1. Causal network under independent and synergistic assumptions. Independent (A; diagonal =FALSE) and synergistic model (B; diagonal = TRUE) yield distinct network architectures. Clinical variables (green) and latent factors (orange) are connected by directed causal edges (solid) or undirected associations (dashed). This abstract is funded by: NHLBI R01HL157879, NCT00608764, NHLBI (U01HL089897 and U01HL089856), NIH contract 75N92023D00011