Viability and cell recovery for shipped PBMC samples
Replicate cryovials from healthy donors (sample set 1) for analysis with recall antigen multimers (ESM Table 1) and replicate cryovials from healthy donors and individuals with type 1 diabetes (sample set 3) for analysis with beta cell and recall antigen multimers (ESM Table 3) were thawed at each participating centre and assessed for viability (Fig. 1). For sample set 1, no differences in recovery were observed among centres (median 64.0%; range 52.5–65.6%), and differences in viability were modest (median 91.5%; range 86.4–95.0%), potentially reflecting a systematic counting bias across centres. For sample set 3, significantly lower viability and recovery rates were obtained in Denver, possibly because the samples were stored for longer before analysis. Despite these differences, each vial provided sufficient material for analysis.
Reproducibility of multimer analysis of recall antigen-reactive CD8+ T cells
To implement the combinatorial assay in each centre, five samples from HLA-A2+ healthy donors (sample set 1) were distributed as identical replicate vials and assessed qualitatively and quantitatively for responses to CMV, Epstein–Barr virus (EBV) and influenza epitopes. All samples were pre-treated with dasatinib and stained using multimer panel 1 (listed in Table 1). Representative data are shown in Fig. 2a. Although some numerical differences were observed (Fig. 2b), the staining results were statistically equivalent among centres for all three viral epitopes, with acceptable %CVs (mean 33.8; SD 12.5) for specificities above the positive threshold of 0.05%. The panel was designed to include five different Qdots (585, 605, 655, 705 and 800) to verify that each centre was able to detect signals in each channel despite the use of different cytometer configurations (ESM Table 4). This sample set also included an HLA-A2− control sample, for which the numbers of positive events detected with recall antigen multimers were consistently lower than or equivalent to those detected with the non-interfacing control multimer in all channels.
Single-centre reproducibility of multimer analysis of beta cell antigen-reactive CD8+ T cells
To validate the detection of beta cell antigen-specific CD8+ T cells, we first evaluated assay reproducibility in a single laboratory (Seattle) by staining three identical replicate vials from five HLA-A2+ individuals with type 1 diabetes (sample set 2). Each blind-coded sample was thawed, pre-treated with dasatinib, stained using multimer panel 2 (listed in Table 1) and analysed before unblinding. As summarised in Table 2, the assay exhibited good agreement between replicate samples based on the observed %CVs. The mean %CV across specificities for beta cell antigen multimers was 18.4. As expected, assay precision was better for viral specificities, with a mean %CV of 6.2. For replicate beta cell epitope-specific frequency measurements, SDs were consistent, ranging from 0.005 to 0.010. As a consequence, %CV values were highest for those specificities with the lowest frequencies in the sample set, such as islet-specific glucose-6-phosphatase catalytic subunit-related protein (IGRP), which yielded a mean %CV of 45.1.
Multi-centre reproducibility of multimer analysis of beta cell antigen-reactive CD8+ T cells
To evaluate assay reproducibility across multiple centres, nine samples from HLA-A2+ individuals with type 1 diabetes, ten samples from HLA-A2+ healthy donors and a clone-spiked sample (sample set 3) were distributed as identical replicate vials in a blinded fashion, pre-treated with dasatinib and stained using multimer panel 2 (listed in Table 1). Staining results for the spiked control and a representative sample are shown in Fig. 3a,b, respectively. For the spiked control sample, each centre observed positive staining (>0.1% of total CD8+ T cells) for the CMV- and PPI-specific clones (Fig. 3c) at percentages close to the theoretical spiked values (0.58% for CMV and 0.34% for PPI vs theoretical values of 0.55% and 0.38%, respectively).
Multimer staining results for each individual sample across all participating centres are shown in Fig. 4. Across all centres, background staining was typically low (with an overall mean of 0.0082% and an SD of 0.01% across all samples), and high positive results tended to be more consistent. To assess the technical reproducibility of the assay, we calculated %CVs across all multi-centre replicates for each multimer specificity (Table 3). The overall mean %CV was 119 for beta cell antigen multimers. As such, the multi-centre beta cell antigen data were 3.5-fold more dispersed than the multi-centre recall antigen data and approximately sixfold more dispersed than the single-centre beta cell antigen data. For replicate beta cell epitope-specific frequency measurements, SDs for the multi-centre data were also higher compared with the single-centre data, ranging from 0.007 to 0.11. %CV values were highest for the PPI epitope, which also had the highest SD, and lowest for the GAD and insulin B epitopes. The decreased precision of the multi-centre data is illustrative of the technical challenges associated with implementing a complex multi-parameter flow assay to perform rare-event analysis using non-identical cytometer configurations.
To further evaluate assay reproducibility across multiple centres, we analysed the data using one-way ANOVA. Systematic variation was not evident for the viral recall antigens or for the beta cell antigens GAD, islet antigen-2 (IA-2), IGRP or islet amyloid polypeptide (IAPP) (Fig. 5). However, significant differences were detected for the insulin B and PPI epitopes (Fig. 6). Specifically, higher frequencies of insulin B-specific CD8+ T cells were detected in Paris compared with Seattle or Leiden, and higher frequencies of PPI-specific CD8+ T cells were detected in Denver compared with all other participating centres. Equivalent differences were not observed for the spiked control sample. As the insulin B and PPI multimer staining data also had the highest SDs, it is likely that systematic error or technical issues contributed to the observed differences between centres.
Central calls were made for each sample before unblinding to assign individuals as having type 1 diabetes or healthy donor based on the presence or absence of multimer-positive populations (>0.05% of viable CD8+ T cells). As summarised in Table 4, these qualitative calls were moderately accurate, with average sensitivity of 68% and specificity of 76% (SDs were 11% and 10%, respectively). The same analysis was repeated in each individual centre (Table 5), but these qualitative calls were less favourable, with average sensitivity of 64% and specificity of 65% (SDs were 16% and 13%, respectively).
Aggregated analysis of beta cell antigen-reactive CD8+ T cells
To evaluate assay results in the context of disease status, multimer staining data from all centres were aggregated (barring data from the spiked sample and PPI-specific frequency measurements from Denver), and pairwise comparisons were performed between individuals with type 1 diabetes and healthy donors. The magnitude of staining with the insulin B and GAD multimers was significantly higher in samples from people with type 1 diabetes compared with samples from healthy donors (ESM Fig. 2). In contrast, no significant differences were observed between groups for PPI, IA-2, IGRP or IAPP. A contingency analysis was also performed by aggregating qualitative calls (ESM Table 5). Samples with GAD multimer-positive populations (>0.05% of viable CD8+ T cells) were significantly enriched in the type 1 diabetes group (21/45 vs 8/50, p = 0.001). Samples with at least one detectable beta cell antigen-reactive CD8+ T cell population were also significantly enriched in the diabetes group (32/45 vs 20/50, p = 0.002). Accordingly, beta cell antigen reactivity within the peripheral CD8+ T cell compartment appears to be more common in individuals with type 1 diabetes relative to a healthy control group.