MAI-T1D
MAI-T1D

A continuous progression axis for type 1 diabetes

Sample-level metadata from HPAP, TEDDY and TRIGR placed on one staging axis, banded by confirmed islet autoantibody count up to diagnosis and by time since onset after it.

352subjects
3,397samples
23assay modalities
3cohorts

Counts are derived from the source metadata workbooks and are reproducible from . Subject identifiers have been replaced with opaque per-cohort labels.

Figure 1

A continuous progression axis for type 1 diabetes, and what three cohorts cover along it

A. One axis, five bands. Position is set by confirmed islet autoantibody count up to diagnosis, and by time since diagnosis after it. Stage 3 is the onset of clinical disease, not a duration — so long-standing disease is a separate band, and the three cohorts turn out to occupy different stretches: TEDDY and TRIGR follow children through seroconversion and stop at or just after diagnosis, while HPAP donors sit a median of 6 years past onset. The three schematic tracks carry the ordering the axis has to reproduce: β-cell stress appears before immune-cell activation, and β-cell mass falls only after both are under way.
B. What changes in the beta cell, and when. Each bar begins where that alteration is first reported and continues as it persists. The order is the point: beta-cell stress is intrinsic and early, altered antigen follows, and immune visibility arrives late — so a sample drawn at Stage 0 and one drawn at Stage 1–2 are not reading the same biology. Our modalities reach these unevenly: the transcriptomes (bulk RNA-seq 103, scRNA-seq 258, scMultiomics 280, snMultiomics 90) cover stress, senescence and prohormone processing; oxygen consumption (87) and perifusion (126) cover mitochondrial and secretory failure; the spatial and cytometry panels (IMC 89, CODEX 91, CyTOF 142, CITE-seq 44, flow 239) cover MHC-I, PD-L1 and infiltration. Neoepitope formation is the one alteration no modality in hand reads directly. Positions are drawn from the published literature on pre-type 1 diabetes beta-cell biology; they are not measured in these cohorts.
C. Coverage is deep at both ends and thin through the middle. Blood is the one compartment all three cohorts share (232 subjects) and is the anchor for cross-cohort alignment. The two sparsest cells are the ones that matter most: only 4 donors carry two or more autoantibodies without a diagnosis, and only 1 donor was sampled within a year of onset — against 64 and 6 subjects in blood at the same two points. That is the interval the progression axis has to bridge.
  • Circle area is proportional to subjects; the count is printed beside it. Dashed circles straddle a band boundary: 27 TRIGR subjects whose autoantibody count at that draw is unknown, and 9 HPAP donors whose disease duration is unrecorded.
  • Stage 1 versus Stage 2 is not separable here — it needs OGTT or fasting glucose, which none of the source workbooks carry. 52 HPAP donors with type 2 diabetes or MODY sit outside the axis and are excluded.
  • TEDDY also contributes 475 stool samples and 107 diet records, which are not tissue compartments. All counts are reproducible from integrate.py.

MAI-T1D Sample Atlas

The 3,397 samples the pseudotime will be built and validated on — HPAP, TEDDY and TRIGR on one staging axis. Filter, then click any sample to see the fields that placed it.

0samples
0subjects
0cohorts
Filters

Stage × genetic risk

Rows are risk tiers; columns are staging bands. Samples whose antibody count is unknown at that moment fall in the narrow seam between two bands.

Disease progression between bands — antibody count unknown at that moment

Matching samples

Click a row to inspect its provenance.

CohortSubjectTimepointModality TissueBandAAbRiskAge (y)
Selected sample
What the axes can and cannot show
Stage 3 is onset, not a duration. The consensus staging stops at diagnosis, so time since onset is a separate band: HPAP donors sit a median of 6 years past diagnosis, while TRIGR's six post-diagnosis draws were all taken within two weeks of it.
Bands 1 and 2 are antibody counts, not glucose stages. Formal Stage 1 versus Stage 2 needs OGTT or fasting glucose, which no source workbook carries. HPAP HbA1c is shown below when it applies; TEDDY and TRIGR have no glucose data at all.
169 samples sit between bands. 54 from TRIGR, which records the antibody count only at seroconversion and at its peak rather than per visit; 115 from 9 HPAP donors whose disease duration is unrecorded, so onset and established disease cannot be told apart. They are drawn in the seam rather than forced into a band.
HPAP has one timepoint per donor. These are organ donors, so an HPAP sample carries the donor's terminal state and cannot move along the axis.
TEDDY has no genetic risk score. Its SNP column is an availability flag set to 1 for every child, so all TEDDY samples sit in the Unknown tier.
Eight TEDDY children carry a one-visit discrepancy. Their earliest per-visit antibody marker is one visit away from the Summary sheet's first-AAB+ confirmation. The per-visit markers were used, since those carry the antibody identities.
Risk tiers are ranked within cohort. HPAP uses PRS tertiles (116 of 194 donors scored); TRIGR uses HLA_risk_group, where 5 means DR4-DQ8 and DR3-DQ2 are both present. The two scales are not interchangeable.