A peripheral blood transcriptomic signature predicts autoantibody development in infants at risk of type 1 diabetes

Abstract

Autoimmune-mediated destruction of pancreatic islet β cells results in type 1 diabetes (T1D). Serum islet autoantibodies usually develop in genetically susceptible individuals in early childhood before T1D onset, with multiple islet autoantibodies predicting diabetes development. However, most at-risk children remain islet-antibody negative, and no test currently identifies those likely to seroconvert. We sought a genomic signature predicting seroconversion risk by integrating longitudinal peripheral blood gene expression profiles collected in high-risk children included in the BABYDIET and DIPP cohorts, of whom 50 seroconverted. Subjects were followed for 10 years to determine time of seroconversion. Any cohort effect and the time of seroconversion were corrected to uncover genes differentially expressed (DE) in seroconverting children. Gene expression signatures associated with seroconversion were evident during the first year of life, with 67 DE genes identified in seroconverting children relative to those remaining antibody negative. These genes contribute to T cell-, DC-, and B cell-related immune responses. Near-birth expression of ADCY9, PTCH1, MEX3B, IL15RA, ZNF714, TENM1, and PLEKHA5, along with HLA risk score predicted seroconversion (AUC 0.85). The ubiquitin-proteasome pathway linked DE genes and T1D susceptibility genes. Therefore, a gene expression signature in infancy predicts risk of seroconversion. Ubiquitination may play a mechanistic role in diabetes progression.

Keywords: Autoimmune diseases; Diabetes; Endocrinology; Immunology.

Figure 1. Integrating longitudinal gene expression data from BABYDIET and DIPP cohorts for differential gene expression analyses.

LOESS fitting–based heatmaps of nonseroconvertors (ABN) and seroconvertors (ABP) after differential gene expression analysis. The point “0 months” in nonseroconvertors represents the mean time of ABP seroconversion and, in seroconvertors, represents the actual time of ABP seroconversion.

Figure 2. Predicting seroconversion using at/near birth gene expression and clinical data.

(A) Number of seroconvertors with at least a single clinic visit by age 1 (n = 25). A cumulative distribution of seroconversion is shown. Step-wise (backward) logistic regression models uncover (B) seroconversion features. Variables that predicted seroconversion were ADCY9, PTCH1, MEX3B, IL15RA, ZNF714, TENM1, PLEKHA5, and HLA score. Receiver operator characteristic curve (ROC), area under ROC (AUC), and Akaike information criterion (AIC) are shown for the prediction model. The shaded circle represents the probability threshold from the logistic regression model (0.233) for a sensitivity and specificity above 80%. (C) Survival curve for the prediction model with 34 subjects stratified as predicted seroconvertors (green) and 58 as predicted nonseroconvertors (red) using the same probability threshold.

Figure 3. Protein-protein interaction network linking the products of differentially expressed genes in blood with T1D susceptibility region candidate genes.

Interactions between T1D susceptibility region gene products obtained through GWAS studies (blue) and differentially expressed gene products (red) are shown.

Figure 4. Protein-protein interaction network linking DE genes with T1D susceptibility region candidate genes identifies NEDD4 as a central hub.

(A) NEDD4 directly interacts with 2 T1D susceptibility genes, ERBB3 and GLIS3. (B) NEDD4 serves as a central hub for 4 (PTCH1, PLEKHA5, MEX3B, ADCY9) of the 7 genes that predict seroconversion.