Poster Presentation The 13th International Congress of the Immunology of Diabetes Society 2013

Identification of unique T1D and T2D sub-groups using unsupervised clustering of longitudinal immune biomarker data (#196)

Johnna D Wesley 1 , Ghanashyam Sarikonda 2 , Ken Coppieters 3 , Matthias von Herrath 1
  1. Novo Nordisk Research Center, Seattle, WA, United States
  2. La Jolla Institute for Allergy & Immunology, San Diego, CA, USA
  3. T1D, Novo Nordisk A/S, Søborg, Denmark

Immune biomarkers (IBMs) will likely play increasingly more critical roles in clinical trials and patient care in the near future. However, IBMs are influenced by variation that is inherent to the individual—independent of therapy—and often undefined. Many clinical trials are cross-sectional designs, rather than longitudinal, that cannot distinguish natural variation from the true treatment effect. Furthermore, analysis of IBM expression and association with disease or therapeutic response is traditionally conducted using supervised clustering methods and directed comparisons. These methods can potentially lead to biased and incomplete conclusions, even in the most rigorous of studies. We collected monthly blood samples from consented adults for one year who were non-diabetic or had been diagnosed with either type 1 or type 2 diabetes (T1D and T2D, respectively). The expression of autoantibodies, the frequency of antigen-specific CD8 T-cells and regulatory T-cells, and the antigen-specific production of IFN- and IL-10 by autoreactive CD4 T-cells were evaluated; clinical data (i.e., HbA1c, insulin usage, and random C-peptide levels) were also collected for T1D subjects. Using a novel unsupervised topological data analysis program, we identified previously unknown subgroups of T1D and T2D subjects who differed significantly with regard to key disease-associated IBMs and disease severity. Though these findings were observed in a small study and additional cohorts needs to be evaluated, the data demonstrate the value of unsupervised clustering in the interrogation of human IBM data and the potential as a patient identification and stratification tool.