Multimodal analysis of the "honeymoon period" in autoimmune diabetes

点击翻译按钮获取中文摘要

The ultimate goal of this proposal is to define composite biomarkers that can be used to improve outcomes in future type 1 diabetes (T1D) clinical trials. T1D is the major cause of diabetes in youth. It is characterized by life-long insulin insufficiency due to autoimmune mediated ß cell destruction. Despite considerable efforts over the past 30+ years, effective therapies are still lacking and there is an urgent need for a cure. Natural history studies indicate that the rate of T1D progression varies greatly between individuals, both before, and after onset. Indeed, the current paucity of validated mechanistic biomarkers that can accurately predict “slow” or “fast” progression is a major impediment to finding a cure. At least 40-60% of patients experience a period of partial remission (PRM) in the first 6 mo after they begin taking insulin. This “honeymoon period” is highly variable, ranging from a few weeks to several years. Like T1D, the factors that govern the onset and duration of PRM are not fully understood. Initially it was believed that PRM is solely a metabolic phenomenon, but there is increasing evidence that the immune system also plays an active part. This leads to the primary hypothesis that underpins our proposal: identification of immunological, metabolic, and demographic features that associate with PRM duration will enable the development of improved clinically actionable composite biomarkers for T1D. Our study has a single specific aim, namely, to define and validate one or more classifiers that can accurately predict fast or slow progression of T1D in the first 2y post-onset from baseline data. This will be achieved through an in depth multimodal analysis of peripheral blood drawn from a cohort of 100 subjects with a recent diagnosis of T1D. A single draw will be made at 3-6 months post diagnosis, and a range of assays performed with DNA, RNA, protein and functional readouts, and ranging in complexity from single analytes to single cell transcriptomes. PRM duration will be determined from clinical data collected over the following 1.5-2y. Subjects will be randomized to training and validation cohorts matched for age, gender, and content of “fast” and “slow” progressors. Features from the analytical data will be used to generate models that predict PRM duration using DIFAcTO, a machine learning algorithm that combines univariate filtering, hierarchical clustering, and LASSO regression, to select non-redundant features that result in an optimal model. Performance of the final models will be evaluated by applying them to the independent validation cohort. The features retained in the resulting models will be prime candidates as composite biomarkers to improve subject stratification at recruitment, and aid identification of responders and non-responders, in future clinical trials. Thus, if successful, our study should have significant impact on the field.