Order Decay in Transcription Regulation in Type 1 Diabetes

摘要

In this study, using type 1 diabetes (T1D) as a model system, we investigate two outstanding challenges in the development of molecular signatures of complex traits: the incorporation of gene interaction structure in signature definition, and the integration of multiple Omics data types to refine signature. The T1D data consists of our previously published transcription profiles in control peripheral blood mononuclear cells (PBMC) induced by sera of 142 human subjects from unrelated healthy controls (uHC), and 3 T1D family cohorts: recent onset (RO-T1D), and healthy siblings of probands that are at high (HRS) or low (LRS) genetic risk for T1D. First both weighted and non-weighted co-expression networks were separately constructed in each cohort and were compared. Several network measures, including edge weight and degree distribution, Shannon's entropy, the λ-coefficient, and h-index, were determined. We found that overall the co-expression networks induced by the RO-T1D cohort are significantly weaker, exhibiting a broad spectrum loss of order and control. More specifically, all T1D family cohorts induced more active and orderly transcription coordination among the innate immunity genes, consistent with our previous report of them sharing a heightened innate inflammatory state. On the other hand, higher coordination of the adaptive immunity genes was only induced by the LRS cohort, potentially explaining their low risk for disease. All the network measures also pointed to the same story, and additionally the importance of the innate immunity genes in determining the transcriptome state of the T1D family cohorts. Next, we integrated the protein-protein interaction (PPI) and the transcriptomic co-expression networks, and focused specifically on the smallest functional units of PPI, the protein complexes (PC). A PC is considered active in a cohort, if its co-expression network is percolated. We found that the RO-T1D cohort activated a significant less number of PC than the others. Overall, whether it is co-expression or protein interaction networks, the four cohorts show striking differences and can be clearly discriminated based on network structural measures. In contrast, gene expression levels alone, without the consideration of underlying interaction networks, could barely differentiate the cohorts. In summary these findings demonstrate the advantage network based metrics in defining molecular signatures.