Integrating Molecular and Metabolomic Markers in T1D Enables Precocious Interventions: Are We Getting There? Free

Type 1 diabetes (T1D) is a chronic autoimmune disorder that typically occurs in genetically predisposed individuals after exposure to environmental triggers, such as nutritional factors or contact with microorganisms (1). The main recognized genetic risk factor is the presence of specific alleles in the major susceptibility locus (HLA) (2). The subsequent subclinical progressive pancreatic islet autoimmunity, which culminates in the destruction of β-cells, is first evidenced by the appearance of islet autoantibodies (3). Efforts to unravel the mechanisms which precede the onset of autoimmunity as well as the development of overt T1D have been vigorously pursued.

The Environmental Determinants of Diabetes in the Young (TEDDY) Consortium has followed 8676 children (922 newborn first-degree relatives of persons with T1D and 7754 children from the general population) with T1D risk HLA-genotypes. Four haplogenotypes for broad HLA diversity were included for the general population (HLA DR3-DQ2/DR4-DQ8, DR4-DQ8/DR4-DQ8, DR4-DQ8/ DR8DQ4, and DR3-DQ2/DR3-DQ2), whereas for the first-degree relatives additional 5 haplogenotypes were included. HLA DRB1*04:03 was an exclusion allele. The TEDDY study has been conducted in the USA, Finland, Germany, and Sweden and has evaluated, besides HLA-genotypes, a number of candidate environmental triggers, including infections, probiotics, micronutrients, and microbiome, in order to identify modifiable environmental factors responsible for the development of pancreatic islet autoimmunity and T1D progression (4).

There is emerging evidence that altered metabolomic profiles are related to pancreatic islet autoantibodies development and overt diabetes. In 2008, Orešič and collaborators reported that the appearance of autoantibodies was preceded by reduced ketoleucine and elevated glutamic acid (5). In 2011, Pflueger and colleagues observed that autoantibody-positive children exhibited higher levels of odd-chain triglycerides and polyunsaturated fatty acid–containing phospholipids than autoantibody-negative children. In addition, children who developed autoantibodies before 2 years of age presented predominantly antibodies against insulin and had lower concentration of methionine compared with those who developed autoantibodies in late childhood (6). Recently, in a part of the TEDDY study, Li et al reported specific metabolomic signatures from children’s plasma after birth until the appearance of the first pancreatic islet autoantibodies. Reduced proline was associated with the appearance of autoantibodies against glutamic acid decarboxylase, whereas reduced branched-chain amino acids, methionine and alanine, as well as fatty acids preceded the appearance of antibodies against insulin. The existence of these distinct metabolic patterns for development of the autoantibodies supports the idea of different initial autoimmunity and consequently different possibilities for primary T1D prevention (7).

A great challenge has been the identification of early molecular, immunological, and metabolic markers that would enable interventions to prevent or delay autoimmunity, affecting the natural history of the disease. In the context to evaluate the predictive nature of metabolites and integrate them with genetics and clinical features, the recent paper of Webb-Robertson and colleagues, published in the Journal of Clinical Endocrinology & Metabolism, titled “Integration of Infant Metabolite, Genetic and Islet Autoimmunity Signatures to Predict Type 1 Diabetes by 6 Years of Age,” is of particular interest. The article describes the development of a model using a combination of genetic, immunological, and time-based metabolites signatures at early infancy (3, 6, and 9 months) to predict the likelihood of developing T1D by the age of 6 years (8).

In the Webb-Robertson work, data were obtained from TEDDY study children (n = 702) either from the general population or from families with T1D with HLA genotypes associated with T1D. Of note, 11.4% children progressed to T1D by the age of 6 years. Their machine learning–based feature selection model utilized 16 features (5 nonmetabolite features such as pancreatic antibodies at 9 months of age, DR3/4, genetic risk scores—polymorphisms associated with T1D, gestational age, and exposure to cow’s milk prior to 6 months of age, 3 metabolites measured at 3 months of age, 5 metabolites measured at 6 months of age, and 3 metabolites measured at 9 months of age). The accuracy of the machine learning model was evaluated by the area under a receiver operating characteristic curve (AUC) and was 0.84, even reducing measurements to 3 and 9 months. The authors identified altered sugar metabolism in infancy as important for progression to T1D by age 6.

Some aspects should be pointed out regarding the paper: the islet autoantibodies were still the strongest contributors to AUC and little difference in AUC could be observed when features beyond autoantibody positivity and HLA antigens, such as metabolomics, were included in the model. Nevertheless, it is highly worthwhile integrating all information on metabolomics to classic risk factors showing slight improvement in the prediction of the progression to T1D in early childhood. Moreover, translating this model to the clinic is still a challenge because it uses complex machine learning techniques and cumbersome metabolomics profiling, which still require independent validation from other laboratories and population. As these tools become increasingly more available, integration of metabolomics data with classical T1D risk factors will allow greater understanding of the disease and the development of precision medicine for individuals with high risk for T1D.

Abbreviations

Abbreviations
 
• AUC

  area under the curve

 
• HLA

  human leukocyte antigen

 
• T1D

  type 1 diabetes

Funding

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Disclosure Statement

The authors have nothing to disclose.

Data Availability

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

References