Time-Resolved Autoantibody Profiling Facilitates Stratification of Preclinical Type 1 Diabetes in Children

Abstract

Progression to clinical type 1 diabetes varies among children who develop β-cell autoantibodies. Differences in autoantibody patterns could relate to disease progression and etiology. Here we modeled complex longitudinal autoantibody profiles by using a novel wavelet-based algorithm. We identified clusters of similar profiles associated with various types of progression among 600 children from The Environmental Determinants of Diabetes in the Young (TEDDY) birth cohort study; these children developed persistent insulin autoantibodies (IAA), GAD autoantibodies (GADA), insulinoma-associated antigen 2 autoantibodies (IA-2A), or a combination of these, and they were followed up prospectively at 3- to 6-month intervals (median follow-up 6.5 years). Children who developed multiple autoantibody types (n = 370) were clustered, and progression from seroconversion to clinical diabetes within 5 years ranged between clusters from 6% (95% CI 0, 17.4) to 84% (59.2, 93.6). Children who seroconverted early in life (median age <2 years) and developed IAA and IA-2A that were stable-positive on follow-up had the highest risk of diabetes, and this risk was unaffected by GADA status. Clusters of children who lacked stable-positive GADA responses contained more boys and lower frequencies of the HLA-DR3 allele. Our novel algorithm allows refined grouping of β-cell autoantibody-positive children who distinctly progressed to clinical type 1 diabetes, and it provides new opportunities in searching for etiological factors and elucidating complex disease mechanisms.

Figure 1

Hierarchical clustering results for longitudinal autoantibody profiles of 370 children who developed multiple β-cell autoantibodies. The dendrogram is divided into 12 multiple autoantibody clusters (mC1–mC12). Each column within a cluster represents the follow-up time from birth (age) for one child. The qualitative status of IAA, GADA, and IA-2A is indicated by color (red = antibody-positive; blue = antibody-negative) with respect to the child’s age when antibodies were measured.

Figure 2

Aggregated longitudinal profiles of IAA, GADA, and IA-2A for children in the multiple-autoantibody clusters (mC1–mC12). For each cluster, the percentages of children who had the respective autoantibodies are indicated by color (white: 0% positive; red: 100% positive) with respect to age. The blue line indicates the age until which >50% of children in the cluster were followed up, and the green lines indicate the age until which >25% of children in the cluster were followed up. Autoantibody profiles are plotted until only two children in the cluster remained in follow-up.

Figure 3

Characteristics of clusters of children who had multiple autoantibodies and who seroconverted at a very young age (median age <2 years). A: The percentage of children in each cluster who were stable-positive, transiently positive, or negative for IAA, GADA, and IA-2A at follow-up. B: The cumulative diabetes-free survival from autoantibody seroconversion. C: The overall frequency of diabetes throughout follow-up.

Figure 4

Progression to type 1 diabetes among clusters of children who were positive for multiple autoantibody types and who have similar autoantibody characteristics but among whom the age at seroconversion varied. Clusters are organized into four groups of three clusters each on the basis of the similarity of autoantibody profiles. A: The percentage of children who were stable-positive, transiently positive, or negative for IAA, GADA, and IA-2A at follow-up. B–E: Cumulative diabetes-free survival from autoantibody seroconversion.

Figure 5

The proportions of boys (A) and HLA-DR genotypes (B) among the children in the multiple-autoantibody clusters (mC1–mC12). Clusters are organized into four groups of three clusters each on the basis of the similarity of autoantibody profiles. The group comprising mC5, mC8, and mC3 included a significantly larger proportion of boys (P = 0.002) and a lower frequency of HLA-DR3 (P = 0.0002) than did the other cluster groups.