Advertisement

14: Methylation differences reveal heterogeneity in spontaneous preterm birth pathophysiology: a visual analytical approach

      Objective

      To determine potential pathways and subphenotypes of spontaneous preterm birth based on genome-wide DNA methylation profiles using visual analytics.

      Study Design

      We analyzed DNA methylation across the genome (HumanMethylation450 BeadChip) in cord blood from 50 African American subjects consisting of 22 cases (24-34 weeks) and 28 controls (>39 weeks). The methylation sites were ranked based on association with gestational age, and the top-10 sites that were located in genes (highly significant after FDR correction) were selected for network analysis. Subject-methylation site relationships were measured using residuals from linear regression, and the data were visualized and analyzed using two bipartite networks: (1) all 50 cases/controls and 10 methylation sites, and (2) only 22 cases and 10 methylation sites. The significance of the results was verified by permutation analysis, and pathways were inferred using Ingenuity Pathway Analysis (IPA).

      Results

      The network with all subjects (Fig. 1A) showed clustering of controls that were strongly hypermethylated at 3 sites compared to the cases, which suggested down regulation of BMI1 (cg23754392) and CDKN2C (cg25592206) resulting in cell cycle arrest and normal physiologic senescence at term. Conversely, the network showed clustering of cases that were hypomethylated at these 3 sites, and strongly hypermethylated at 7 other sites. Of these, hypermethylation of CpG sites in BCL9 (cg10020892), and IRF8 (cg 16705546), which are cell cycle promoters, suggests that senescence is an unlikely pathology in PTB in the absence of preterm premature rupture of membranes. Furthermore, a case-only analysis (Fig. 1B) revealed that a subset of the cases have hypomethylated CpG sites in BMI1, CDKN2C, and IRF8, suggesting heterogeneity in PTB pathophysiology.

      Conclusion

      Bipartite network analysis helped to reveal heterogeneity in PTB and to infer the pathways involved in subphenotypes. These results should help improve the future modeling of PTB risk.
      Figure thumbnail fx1
      (A) Bipartite network of 50 cases/controls, and top-10 significant methylation sites, with superimposed clusters shaded in blue. (B) Bipartite network of only 22 preterm cases, and top-10 significant methylation sites with superimposed clusters, revealing 2 subphenotypes.