Objective
This proposal addresses the RFA “Improving Risk Prediction for T1D Onset: Modeling and Assay Development for Clinical and Commercial Use”. The aim of the RFA is to improve risk prediction of T1D onset in two areas: in silico modeling using existing genetic data to improve identification of the at-risk population, and integration of T1D genetics or autoantibody testing into assay platforms that are commercially available as direct to consumer products or routinely used in clinical/public health laboratories. In response to the RFA, the current proposal encompasses both of the objectives and represents a collaboration at the University of Virginia between Dr. Farhy in the Center for Diabetes Technology (CDT) and Dr. Rich in the Center for Public Health Genomics (CPHG). At the request of JDRF, the two components are in this one application.
Currently, lab hospital tests for presence of islet autoantibodies are used to detect progression to T1D, typically in siblings of those living with T1D; unfortunately, up to 95% of people who develop T1D do not have a family history of the disease. This absence of a family history means that the signs and symptoms of T1D are often missed, and this lack of immediate care and monitoring has led to the increase in diabetic ketoacidosis (DKA) in up to 40% of those who develop T1D. Thus, there is urgent need for increased predictive tests that are available at home and not requiring hospitalization.
The first objective of our proposal is to develop a self-administered home test, based on the use of Continuous Glucose Monitoring (CGM) that measures the response to a standardized mixed meal (breakfast) over a one-week period. We have developed methods for analysis of CGM data that are predictive of development of T1D, yet we believe that the combination of machine learning/artificial intelligence methods for analysis of the collected data will improve predictive performance. With existing data (and data currently being collected), we will be able to use the in-home CGM data and incorporate genetic risk data (also collected in-home) that is predictive of the risk for progression to T1D without visiting a clinical laboratory or hospital.
The second objective of our project is to use existing whole genome sequencing data from individuals with T1D and controls, of diverse (African and Hispanic/Latinx) genetic ancestry to improve existing genetic risk scores (GRS) and develop a new polygenic risk score (PRS). T1D risk is ~50% due to genetic factors. We have identified ~85% of the genetic risk of T1D, but these studies have been in individuals of European Caucasian (white) ancestry. We have assembled the largest collection of whole genome sequence data in subjects with type 1 diabetes and in controls of African and Hispanic/Latino (as well as European) ancestry for detection of novel genetic risk factors, refinement of existing risk regions for detection of causal variants, and establishment of improved genetic/polygenic risk scores to enable detection of T1D risk in individuals of diverse genetic ancestry and eligible for islet autoantibody screening and entry into immune therapy trials.
Background Rationale
This proposed research from two teams at the University of Virginia addresses the components of the JDRF RFA on improving risk prediction for T1D onset using existing data. Type 1 diabetes (T1D) is an autoimmune disease in which unknown environmental factors cause the body’s immune system to attack and destroy the insulin-producing beta cells in the pancreas. From twin and family studies, the risk of T1D is due, almost equally, from genetic factors and non-genetic (environmental) factors. Although there is strong support for genetics in risk of T1D, only 5% of those who develop T1D have a family history of the disease. Fortunately, most of the genetic risk of T1D has been identified (led by our group), but almost exclusively in populations of European Caucasian (white) genetic ancestry. While these studies have permitted this knowledge to develop genetic risk scores (GRS) that can identify those at “high genetic risk” of T1D, it is based upon limited genomic information (only a small fraction of the human genome has been covered) and is restricted to white genetic ancestry. A major limitation is that the transfer of a T1D GRS to other, non-white populations does not perform well in detecting those at high genetic risk.
Many studies have shown that screening individuals who are at increased risk by being a sibling of someone with disease can reduce risk of diabetic ketoacidosis (DKA), a severe complication at clinical onset of T1D. As the rate of DKA at onset is nearly 40% in those who are not monitored, the detection of those at high risk without a family history of T1D is critical. Testing for presence of islet autoantibodies has several limitations, including questions of whom to test and, given the timing of appearance of specific islet autoantibodies can vary, and when to test has been a matter of debate. Further, testing for islet autoantibodies requires a visit to a clinical laboratory or hospital.
Recent research have shown that analysis of continuous glucose monitoring (CGM) data can be used as an alternative to predict the risk for progression to T1D. These efforts utilize traditional CGM aggregated metrics which cannot capture the richness of the CGM information. Our group has developed multiple advanced methods for analysis of CGM data and has designed a unique self-administered home test which, in healthy relatives of individuals with type 1 diabetes, can distinguish between different immunological risk (number of islet autoantibodies). In this application, we will first use existing data collected during this dedicated self-administered CGM home test in relatives of individuals with T1D (but themselves without T1D), as well as data being generated in a separate in-home study. We will develop a new, improved machine learning/artificial intelligence-based method to improve prediction for simultaneous assessment of the immunological and genetic risks for an individual to develop T1D. Second, we will conduct analysis of our existing whole genome sequence data on T1D cases and controls of African, Hispanic/Latino, and European ancestry, a sample that represents the largest number of individuals with whole genome sequencing and of diverse ancestry in T1D to date. The analysis of these data will provide novel information on genetic risk of T1D across the whole genome and develop a “global” polygenic risk score to identify genetically at-risk individuals who would be eligible for autoantibody screening and entry into immune intervention trials.
Description of Project
Type 1 diabetes (T1D) is an autoimmune disease in which the body’s immune system attacks and destroys the insulin-producing beta cells in the pancreas in response to an unknown environmental trigger. T1D occurs in up to 1 in 300 children, the vast majority (~95%) not having a family history of the disease. Nearly one-half of the risk of type 1 diabetes is due to genetic factors with the remainder from unknown environmental triggers (e.g., viruses).
We have now identified ~85% of the genetic risk and developed genetic risk scores (GRS) that help predict those children at genetic risk and eligible for islet autoantibody screening and immune intervention. Major limitations of prediction come from the current T1D GRS that are based upon European Caucasian (white) ancestry and a limited coverage of the human genome. In one part of this project, we will utilize a unique collection of data obtained by whole genome sequencing generated by our group on 9,363 participants from four studies – the Type 1 Diabetes Genetics Consortium (T1DGC, n=3,458, cases and controls), The Environmental Determinants of Diabetes in the Young (TEDDY, n=1,125, cases and controls), the Diabetes Auto Immunity Study in the Young (DAISY, n=160, cases and controls), and the Multi-Ethnic Study of Atherosclerosis (MESA, n=4,620, all controls). These studies provide whole genome sequencing data on subjects of African, Hispanic/Latino, and European ancestry to identify novel genetic risk variants and improve prediction of T1D GRS across multi-ethnic populations.
Many studies have shown that timely screening for the risk of T1D can be used to lessen multiple problems occurring in both the early and late stages of disease development, and will improve health outcomes. For individuals at genetic risk of T1D, testing for presence of islet autoantibodies is used to determine the stage of progression to T1D; however, the timing of their appearance, and when to test, has been a matter of debate. Recently, it was shown that continuous glucose monitoring (CGM) can be used as an alternative to predict progression to T1D. Our group has developed multiple advanced analytical methods for analysis of CGM data, including a unique self-administered home test that can distinguish between people at different immunological risk (number of islet autoantibodies). In the other part of this project, we will use existing CGM data from 60 healthy relatives of individuals with T1D obtained from an NIH-funded TrialNet ancillary study from the self-administered home test. We will develop an improved machine-learning/artificial intelligence analysis of these CGM data, with validation in 100 individuals in an ongoing study supported by the Helmsley Charitable Trust that integrates genetic, immunologic and CGM-based risk assessment for development and glycemic control in T1D.
With this proposal, we will improve existing genetic prediction of T1D risk across diverse ancestries, develop improved genetic and polygenic risk scores, and validate self-administered in-home CGM test technology for simultaneous assessment of the T1D risks that can be implemented into population-level assessment.
Anticipated Outcome
This project requests support from JDRF to use existing data for improvement of prediction of T1D. The project consists of two parts.
The first part uses existing data from an NIH-funded study of an in-home CGM-based response to a standard mixed-meal breakfast (taken three times in a 7-day period). Using advanced analytic methods, the data from the in-home test with CGM improves prediction of T1D risk in people without T1D but with a relative living with T1D. These data from the self-administered, home test will be mined using advanced machine learning/artificial intelligence data analysis methods to improve the prediction of risk for an individual to develop T1D. For confirmation of the improvement in prediction, an independent set of data, now collecting in-home CGM-based response to the mixed meal breakfasts using the same protocol (supported by the Helmsley Charitable Trust) will be used. This assessment can be incorporated into practice as an in-home risk-assessment technology that does not require a laboratory or hospital visit, as the CGM data can be accessed and analyzed remotely.
The second part uses existing whole genome sequence data obtained from four different sources. The University of Virginia group (Dr Rich) has served as the lead in the Type 1 Diabetes Genetics Consortium (T1DGC) that has produced whole genome sequence data in T1D cases and controls of African, Hispanic/Latinx and European ancestry; in the TEDDY and DAISY studies with European and Hispanic/Latinx ancestry in T1D cases, islet autoantibody (but not clinical T1D) positive individuals, and controls; and in the Multi-Ethnic Study of Atherosclerosis (MESA) that serve as controls. In total, over 9,000 subjects will be included in the whole genome sequencing analysis. From these data, the genetic variation in the human genome in diverse ancestry populations with T1D will be analyzed for discovery of novel sites in the genome associate with risk, age at onset, and used for the development of both genetic and polygenic risk scores that can be applied to global populations.
Together, this project addresses two aspects of prediction of T1D risk and addresses major gaps in accessibility and diversity in T1D research.
Relevance to T1D
This proposal addresses two gaps in knowledge and implementation to improve prediction and identification of those at risk of T1D. The application for support from JDRF consists of two areas of relevance to T1D, with both being implemented as in-home tests and data acquisition.
Current immune intervention trials have focused on unaffected relatives of individuals with T1D for inclusion in the trials who have a baseline risk of 8% (above the population rate, 0.5%) but also have islet autoantibodies. The presence of islet autoantibodies is evidence of the autoimmune process that leads to beta cell destruction and development of T1D, with a projected risk of T1D approaching 80% over 5-8 years. Unlike genetic information, which remains constant over time, the process of autoimmunity has uncertain start time or progression, requiring continuous testing. A major gap is providing technologies that can be used at home, rather than in a specialty clinic or hospital, and yielding data that are highly predictive of T1D.
Our group has pioneered the development of CGM-based methodologies to predict T1D risk using an in-home, 7-day protocol that uses a standardized mixed-meal breakfast that captures how well the meal generates physiologic response to the meal. The CGM data currently exhibits excellent prediction of risk; however, improvements in prediction can be made by application of machine learning/artificial intelligence methods. The availability of immunological and genetic risks for T1D can be incorporated with these data to further increase predictive ability.
In addition, our group has led the discovery of genetic variation associated with T1D risk through the Type 1 Diabetes Genetics Consortium (T1DGC), the TEDDY study, DAISY, and others. We have identified ~85% of the genetic risk in populations of European Caucasian (white) ancestry, but little has been done in non-white populations. Also, the current information in European Caucasians is not in-depth, but based upon genotyping array and fine-mapping approaches. Thus, we propose to address this gap by analysis of whole genome sequencing data in over 9,000 subjects of African, Hispanic/Latinx and European ancestry with T1D, islet autoimmunity and controls. Further, we will generate ancestry-specific genetic risk scores (GRS) and polygenic risk scores (PRS) that will enhance the identification of those in diverse ancestry populations at high genetic risk, eligible for autoantibody screening, and enrollment into immune intervention trials.
Together, this project will address gaps in knowledge to improve prediction of risk of T1D, using new methodologies and metrics derived from advanced machine learning analysis of a dedicated CGM home test and the analyses of whole genome sequence data on T1D in diverse populations. The results as applied to these existing data are highly relevant to the mission of the JDRF and will improve and facilitate screening and assessment of the risk for progression to T1D and help identify subjects for future prevention trials.