Dysregulation of blood glucose and triglycerides are the major characteristics of type 2 diabetes mellitus. We sought to identify the mechanisms regulating blood glucose and lipid homeostasis. Cell-based studies established that the Foxo forkhead transcription factors Forkhead box O (Foxo)-1, Foxo3, and Foxo4 are inactivated by insulin via a phosphatidylinositol 3-kinase/Akt-dependent pathway, but the role of Foxo transcription factors in the liver in regulating nutrient metabolism is incompletely understood. In this study, we used the Cre/LoxP genetic approach to delete the Foxo1, Foxo3, and Foxo4 genes individually or a combination of two or all in the liver of lean or db/db mice and assessed the role of Foxo inactivation in regulating glucose and lipid homeostasis in vivo. In the lean mice or db/db mice, hepatic deletion of Foxo1, rather than Foxo3 or Foxo4, caused a modest reduction in blood glucose concentrations and barely affected lipid homeostasis. Combined deletion of Foxo1 and Foxo3 decreased blood glucose levels, elevated serum triglyceride and cholesterol concentrations, and increased hepatic lipid secretion and caused hepatosteatosis. Analysis of the liver transcripts established a prominent role of Foxo1 in regulating gene expression of gluconeogenic enzymes and Foxo3 in the expression of lipogenic enzymes. Our findings indicate that Foxo1 and Foxo3 inactivation serves as a potential mechanism by which insulin reduces hepatic glucose production and increases hepatic lipid synthesis and secretion in healthy and diabetic states.

Cardiac failure is a major cause of death in patients with type 2 diabetes, but the molecular mechanism that links diabetes to heart failure remains unclear. Insulin resistance is a hallmark of type 2 diabetes, and insulin receptor substrates 1 and 2 (IRS1 and IRS2) are the major insulin-signaling components regulating cellular metabolism and survival. To determine the role of IRS1 and IRS2 in the heart and examine whether hyperinsulinemia causes myocardial insulin resistance and cellular dysfunction via IRS1 and IRS2, we generated heart-specific IRS1 and IRS2 gene double-knockout (H-DKO) mice and liver-specific IRS1 and IRS2 double-knockout (L-DKO) mice. H-DKO mice had reduced ventricular mass; developed cardiac apoptosis, fibrosis, and failure; and showed diminished Akt→forkhead box class O-1 signaling that was accompanied by impaired cardiac metabolic gene expression and reduced ATP content. L-DKO mice had decreased cardiac IRS1 and IRS2 proteins and exhibited features of heart failure, with impaired cardiac energy metabolism gene expression and activation of p38α mitogen-activated protein kinase (p38). Using neonatal rat ventricular cardiomyocytes, we further found that chronic insulin exposure reduced IRS1 and IRS2 proteins and prevented insulin action through activation of p38, revealing a fundamental mechanism of cardiac dysfunction during insulin resistance and type 2 diabetes.

The past decade has seen a revolution in genomic technologies that enabled a flood of genome-wide profiling of chromatin marks. Recent literature tried to understand gene regulation by predicting gene expression from large-scale chromatin measurements. Two fundamental challenges exist for such learning tasks: (1) genome-wide chromatin signals are spatially structured, high-dimensional and highly modular; and (2) the core aim is to understand what the relevant factors are and how they work together. Previous studies either failed to model complex dependencies among input signals or relied on separate feature analysis to explain the decisions. This paper presents an attention-based deep learning approach, AttentiveChrome, that uses a unified architecture to model and to interpret dependencies among chromatin factors for controlling gene regulation. AttentiveChrome uses a hierarchy of multiple Long Short-Term Memory (LSTM) modules to encode the input signals and to model how various chromatin marks cooperate automatically. AttentiveChrome trains two levels of attention jointly with the target prediction, enabling it to attend differentially to relevant marks and to locate important positions per mark. We evaluate the model across 56 different cell types (tasks) in humans. Not only is the proposed architecture more accurate, but its attention scores provide a better interpretation than state-of-the-art feature visualization methods such as saliency maps.1