The specimens used in this study were obtained from the DPT-1 trial cohort sponsored by the NIDDK between 1994 and 2003 (18). Detailed patient recruitment and study protocols have been reported previously (18). Briefly, all participants were first- or second-degree of relatives of a person with T1D and were tested for islet cell autoantibodies (ICAs). Written informed consent was obtained from all the subjects in the study group. Patients with ICA autoantibodies were offered additional testing for GAD, IA-2, and insulin autoantibodies. Islet autoantibody testing records, follow-up records, and clinical diagnosis of stage 3 T1D records were available from the NIDDK biorepository.

Sera collected within 6 months of study enrollment were obtained from a total of 48 subjects who progressed to stage 3 T1D and 44 subjects who did not progress to stage 3 T1D during the follow-up. The subjects were randomly selected by the NIDDK central repository staff. These subjects either developed stage 3 T1D during follow-up or were followed up for at least 5 years. The demographic characteristics of the study participants are presented in Table 1 . Notably, the study participants were predominantly non-Hispanic white individuals. There were more male than female participants. The samples were transferred to Enable Biosciences for multiplex ADAP analysis as de-identified-coded specimens. The result was only unblinded by the NIDDK central repository after testing was completed. The study was approved by the Western IRB (IRB number #20180015) to Enable Biosciences.

Previously, we reported a multiplex ADAP method for detecting three islet autoantibodies (15). In addition, we described an automated Hamilton MicroLab STAR system to carry out the 3-plex ADAP assay (16). Recently, we expanded the assay to 5-plex to test for IAA, GADA, IA2A, ZnT8A, and TGA on a modified version of Hamilton MicroLab STAR to achieve full automation (17). Herein, we restricted the ADAP assay to a 4-plex assay to test for all four islet autoantibodies (IAA, GADA, IA2A, and ZnT8A) on the Hamilton MicroLab STAR system. Briefly, 4 μL of serum was incubated with 8 μL of DNA-barcoded autoantigens at 37°C for 30 min. If present in the specimens, autoantibodies agglutinate autoantigens into a dense immune complex. Then, 4 μL of the mixture was aspired and mixed with 116 μL of ligation mixtures, where nearby DNA in the dense immune complex was ligated to form a full-length DNA amplicon. Next, 25 μL of the above mixture was further mixed with 25 μL of PCR amplification mixtures containing primers for all five autoantibodies for a total of 13 PCR cycles using an on-deck thermocycler (ODTC, Inheco, Martinsried, Germany). The amplified products were then aspired to 384 well plates in which each well contained the cognate primer pairs for each autoantibody to achieve specific quantification by real-time quantitative PCR (RT-qPCR). The qPCR-ready plates were transferred to Bio-Rad CFX384 to enable an automated sample-to-answer solution. The samples were analyzed in a coded and randomized manner. The results were unblinded after sample testing was completed. The assay cutoffs were determined by testing 80 healthy controls and set at the 99th percentile. The cut-offs for IAA, GADA, IA2A, and ZnT8 were 0.99, 3.1, 2.3, and 2.0, respectively.

The GAD, IA-2, and insulin autoantibody testing results were obtained from the NIDDK central repository database. Laboratory procedures for GAD, IA-2, and insulin autoantibody analyses have been extensively reported (18). Briefly, GAD and IA-2 autoantibodies were detetcted at the Barbara Davis Center (Denver, CO, USA). Insulin autoantibody levels were determined at the Barbara Davis Center or Joslin Diabetes Center (Boston, MA, USA). The cut-off values for the GAD and IA-2 assays were 0.032 and 0.049, respectively. For the insulin assays, the cut-off was 0.01 at the Barbara Davis Center and 0.02 at Joslin Diabetes Center. The cutoffs were determined using the 99th percentile of the healthy controls. A combined radiobinding assay was performed for GAD and IA-2 autoantibodies using radioactively labeled H3-GAD65 and S35-IA-2.

Positive predictive value (PPV) was defined as the probability that a subject with a positive test result actually progressed to clinical presentation of the disease. The negative predictive value (NPV) was defined as the probability that a subject with a negative test result truly did not progress to disease clinical presentation. For instance, in this study, a positive test result was defined as having two or more islet autoantibodies, unless otherwise noted. The overall PPV was calculated based on the number of individuals that progressed to stage 3 T1D during the entire follow-up period, while the overall NPV was calculated based on the number of individuals who did not progress to stage 3 T1D during the entire follow-up period. The 5-year risk PPV and NPV were calculated similarly, except that we restricted the analysis to progression within 5 years. It should be noted that all study subjects had either been followed for 5 years or progressed to stage 3 T1D within 5 years. Kaplan–Meier estimates were used to plot progression risk and to compare probabilities of stage 3 T1D progression in subjects stratified by the number of islet autoantibodies, sex, or age groups. For all analyses, a 2-tailed P-value of 0.05 was considered significant. All statistical analyses were performed using Graphpad Prism (version 9.3.1).

All data generated or analyzed during this study are included in the published article (and its online Supplementary Files ). The reagents used in this study are available from the corresponding author upon request.

In this study, we obtained 92 sera samples from 48 progressors and 44 non-progressors in the NIDDK DPT-1 study (18). All individuals either developed T1D during the follow-up period (progressors) or were followed up for at least 5 years (non-progressors). The sera were analyzed using multiplex ADAP assays for autoantibodies against GAD, IA-2, insulin, and ZnT8 ( Figure 1 , Table 1 ). Among them, 68 individuals tested positive for two or more islet autoantibodies, and 46 developed stage 3 T1D during the follow-up period. The median time from positivity for two or more islet autoantibodies to stage 3 T1D diagnosis was 4.2 years (Range: 1.0–8.4 years). Among the 24 individuals with one or fewer islet autoantibodies, only two individuals progressed to stage 3 T1D. One of them, diagnosed at the age of 13.1 years old, had a high level of GAD autoantibody and IA2 autoantibody level immediately below the cut-off, while the other, diagnosed at age of 27.8 years old, was negative for all islet autoantibodies. The overall positive predictive value (PPV) and negative predictive value (NPV) of the multiplex ADAP islet autoantibody assay based on the presence of two or more islet autoantibodies were 68% (46/68) and 92% (22/24), respectively. Alternatively, the PPV and NPV for progression to stage 3 T1D within 5 years of testing were 49% and 92%, respectively. The 5-year PPV was lower than the overall PPV because some individuals developed stage 3 T1D after 5-years of initial testing. The 5-year PPV observed in this study is consistent with that of other longitudinal follow-up studies (19–21).

Next, we sought to further explore whether individuals with two, three, or four islet autoantibodies would have distinct progression risks to stage 3 T1D ( Figure 1B ). Progression rates ranged from 64% to 70% ( Supplementary Table 1 ). The median time from positivity for two or more islet autoantibodies to clinical presentation was 3.9, 3.1, and 4.4 years for individuals with two, three, and four islet autoantibodies, respectively.

Furthermore, an analysis was conducted to evaluate whether the types of islet autoantibodies would influence the risks of progression to stage 3 T1D ( Tables 2 , 3 ). For individuals with two or more islet autoantibodies, if their autoantibody positivity included GAD, IA-2, insulin, or ZnT8 autoantibodies, the median time to diagnosis was 3.7, 3.8, 3.6, and 3.6 years, respectively, and the PPV were 68%, 68%, 65%, and 71%, respectively. If the autoantibody positivity included GAD/IA-2, GAD/insulin, GAD/ZnT8, IA-2/insulin, IA-2/ZnT8, Insulin/ZnT8 autoantibodies, the median time to diagnosis was 3.8, 3.6, 3.6, 3.8, 3.6, and 4.2 years, respectively and the PPV was 68%, 65%, 71%, 64%, 73%, and 67%, respectively.

Therefore, the above observation indicated that the presence of multiple islet autoantibodies was a critical risk factor for progression to stage 3 T1D.

Patients positive for two or more islet autoantibodies might have distinct progression risks depending on their age and sex (22). To investigate this further, we first stratified the individuals into those under and above the age of 8 at the time of testing. For individuals under the age of 8 years, the multiplex ADAP assay had a PPV and NPV of 80% and 100%, respectively. For individuals over age of 8 years, the PPV and NPV were 58% and 90%, respectively. Therefore, the development of multiple islet autoantibodies at a young age appears to increase the risk of progression risk to stage 3 T1D. On the other hand, female patients had a slightly higher PPV than male patients (70% vs 67%), and the NPV was comparable (92% vs 91%).

Recently, ZnT8 autoantibodies were discovered. The value of ZnT8 autoantibodies in aiding the diagnosis of new-onset clinical diabetes and risk predictions has been widely reported (23). It is of great interest to investigate whether the exclusion of ZnT8 autoantibodies would substantially impact the prediction of stage 3 T1D progression risk.

To this end, the above analysis was performed again using only GAD, IA-2, and insulin autoantibodies ( Figure 2 ). Intriguingly, 67 individuals tested positive for two or more islet autoantibodies, and 46 out of the 67 individuals eventually developed stage 3 T1D during follow-up. The median time from positivity for two or more islet autoantibodies to clinical presentations was 3.7 years. Similar to the previous analysis, only two out of 25 individuals with one or fewer islet autoantibodies progressed to stage 3 T1D. Accordingly, the positive predictive value (PPV) of the multiplex ADAP islet autoantibody assay with GAD, IA-2, and insulin autoantibodies was 68% (46/67) and the negative predictive value (NPV) was 92% (23/25). These predictive values were statistically indistinguishable from the predictive values when all four islet autoantibodies were included. The data thus support the use of three cardinal islet autoantibodies for the prediction of the risk of progression to stage 3 T1D.

The prediction of the risk of progression to stage 3 T1D has been extensively studied in several landmark studies using radiobinding assays to measure islet autoantibodies. Indeed, the underlying DPT-1 study was one of the earliest nationwide longitudinal studies to provide critical insight into the natural history of T1D development and inspired and shaped study designs for many other recent studies. Importantly, radiobinding assay data from the DPT-1 studies were available from the NIDDK biorepository. We sought to compare the risk prediction between the multiplex ADAP assays and radiobinding assays. It should be noted that DPT-1 study was conducted between 1994 and 2003 (18). The design and protocols for radiobinding assays have been improved in recent studies (10). Nevertheless, the data will provide a valuable context to help understand whether the observed multiplex ADAP assay performance is satisfactory.

Among the 92 patients with radiobinding assay data for GAD, IA-2, and insulin autoantibodies, 51 tested positive for two or more islet autoantibodies, and 34 progressed to stage 3 T1D during the follow-up period, with a median time to diagnosis of 3.4 years ( Figure 3 , Table 4 ). Fourteen of the 41 individuals with one or no islet autoantibodies progressed to stage 3 T1D, with a median time to diagnosis of 4.3 years. The overall PPV and NPV of the radiobinding assays were 67% and 66%, respectively. The 5-year PPV and NPV for the radiobinding assays were 51% and 78%, respectively.

To compare performance, we first restricted the multiplex ADAP assay analysis to GAD, IA-2, and insulin autoantibodies, given that ZnT8 autoantibodies were not yet discovered at the time of the DPT-1 study. The multiplex ADAP assay and radiobinding assays had similar PPV of 68% and 67%, respectively. Nevertheless, the NPV differences were statistically significant (92% vs 66%, respectively). To elucidate the potential sources of the NPV differences, it was noted that multiplex ADAP assays identified 46 out of 48 patients that progressed to stage 3 T1D as multiple islet autoantibody-positive. In contrast, radiobinding assays only identified 34 out of 48 progressors as multiple islet autoantibody-positive, leading to a lower NPV.

We further compared the pattern of islet autoantibodies for the 12 progressors that had discrepant assigned risk profiles using multiplex ADAP and radiobinding assays ( Table 5 ). Five of the 12 progressors were positive for GAD/IA-2/insulin autoantibodies, and the remaining seven individuals were positive for GAD/IA-2 autoantibodies with multiplex ADAP assays. On the other hand, seven out of the 12 progressors were single positive for GAD autoantibodies, one out of 12 was single positive for IA-2 autoantibodies, one out of 12 was single positive for insulin autoantibodies, and three out of 12 were negative for all islet autoantibodies by radiobinding assays. Thus, it appears that the discrepant risk profiles were not a result of specific islet autoantibodies. Nevertheless, it was noted that for the seven individuals with single GAD autoantibodies by radiobinding assays, their GAD autoantobody signals measured by ADAP ranged from 7.59 to 11.63, whereas those five were missed by radiobinding assays, and their GAD autoantibody signals measured by ADAP ranged from 4.79 to 6.54. Similarly, for the one individual with insulin autoantibodies by radiobinding assays, the ADAP signal was 4.92, while rest of 4 ADAP insulin autoantibody-positive individuals had signals from 1.07 to 2.07. These observations suggested that ADAP had improved sensitivities over radiobinding assays for GAD and insulin autoantibodies, as the samples were only radiobinding assay-positive if their ADAP signals were higher in values. In contrast, for IA-2 autoantibodies, the only radiobinding assay-positive sample had an ADAP signal of 7.78, but the remaining 11 samples had ADAP signals from 2.42 to 11.64. Should sensitivities be the only factor, we would expect those samples with ADAP signals above 7.78 to be positive by radiobinding assays. The fact that several samples with strong ADAP signals were negative by radiobinding assays implied that the two assays might have additional differences for IA-2 autoantibody detection, such as autoantibody epitopes and isotypes.

Notably, the NIDDK biorepository had longitudinal radiobinding assay data for a portion of DPT-1 study samples. For these 12 progressors who were initially positive for one or fewer islet autoantibodies by radiobinding assays, five later developed two or more islet autoantibodies. The ADAP assay preceded the radiobinding assay by a median of 2.8 years for detecting two or more islet autoantibodies in these five samples. The remaining seven progressors did not develop two or more islet autoantibodies by radiobinding assays during the follow-up. While the sample size was limited, this is preliminary evidence that the ADAP assay could enable earlier diagnosis of stage 1 or stage 2 T1D.

The overall sensitivity of the multiplex ADAP sand radiobinding assay was 96% and 71%, respectively, whereas the overall specificity of the multiplex ADAP sand radiobinding assay was 50% and 61%, respectively.