TOL-3021 SUNRISE: a proinsulin DNA vaccine ran out of biological room

What was tried

Tolerion ran SUNRISE (NCT03895437) as a 78-participant, multicenter, double-blind, placebo-controlled Phase 2 study of TOL-3021, the renamed second-generation form of BHT-3021. The agent is a bacterial plasmid expression vector carrying the coding sequence for full-length human proinsulin (INS, ENSG00000254647). Eligible participants were aged 12 to less than 41 with type 1 diabetes diagnosed within the prior five years, stratified by disease duration. Dosing was 1 mg of plasmid administered intramuscularly weekly for 12 weeks, then monthly to week 52, with a two-year safety follow-up. The primary endpoint was the treatment effect on log-transformed mixed-meal tolerance test C-peptide area under the curve at 12, 16, and 24 weeks, the standard preservation-of-residual-beta-cell-function readout used by TrialNet and FDA reviewers. The trial was terminated on 21 December 2020 when the data safety monitoring board concluded there was a low probability the study would meet its primary endpoint as designed. The status update was posted to ClinicalTrials.gov on 14 May 2026.

The biological hypothesis

TOL-3021 was built on the hypothesis that antigen-specific tolerance could be induced in established T1D by repeatedly presenting full-length proinsulin in a tolerogenic context. The construct uses a methylation-modified backbone that lowers CpG content and is expressed in a non-inflammatory milieu. Resident antigen-presenting cells were expected to prime regulatory T cells specific to proinsulin epitopes rather than activate cytotoxic responses. The pilot study (Roep et al., Sci Transl Med 2013) reported a reduction in proinsulin-reactive CD8+ T cells in peripheral blood with stable C-peptide over 15 weeks at the 1 mg dose, framed as proof of mechanism. Genetic support for INS-directed therapy is high. Open Targets overall association score for INS with type 1 diabetes is 0.690, and the INS VNTR region represents one of the two strongest non-HLA T1D susceptibility loci (Pugliese et al., Nat Genet 1997).

What actually happened

The DSMB stopped enrollment at 78 randomized participants for futility. Trial-level outcomes have not been published, and per the ClinicalTrials.gov record results are not posted. The mechanism failure is therefore inferential rather than directly readable, but two observations constrain the interpretation. The pilot effect (Roep 2013) was demonstrated at 15 weeks in 80 participants with a between-group C-peptide difference that did not survive the 12-month and 24-month follow-ups in the original cohort, where placebo-arm decline caught up. The SUNRISE primary endpoint integrated C-peptide AUC across weeks 12, 16, and 24, exactly the window where the pilot's nominal effect appeared and then collapsed. The DSMB call is consistent with a class effect. Antigen-specific therapies for established T1D have repeatedly produced transient immunological signatures without preserving C-peptide on the standard preservation timeline.

Failure mechanism, best guess

The most parsimonious mechanism is timing mismatch between intervention and beta cell trajectory. By the time mixed-meal C-peptide AUC is measurable in new-onset T1D, autoreactive memory CD8+ T cells specific to proinsulin and other islet antigens have already diversified beyond a single epitope (Culina et al., Sci Immunol 2018,, and Bender et al., Nature 2022). Inducing peripheral tolerance to a single antigen at that stage faces a polyclonal effector pool that does not need INS-specific licensing to continue killing beta cells. Teplizumab succeeds in at-risk stage 2 disease (Herold et al., NEJM 2019) because it acts before this diversification stabilizes. In established disease the same anti-CD3 approach buys less time. TOL-3021 also operated entirely upstream of glucose-responsive insulin secretion. Any preserved C-peptide would have to come from beta cells whose effector burden was already established. The proinsulin DNA platform did not appear to drive new autoimmunity, which is reassuring biologically, but it also did not generate the magnitude of Treg expansion needed to slow disease in the SUNRISE window.

How to prevent this next time

Two quantitative levers would have shifted the design.

First, a Bayesian stopping framework anchored to the pilot's effect size rather than to the protocol's binary 12-week endpoint would have flagged futility earlier and at lower cost. Using the SUNRISE design assumption (relative C-peptide preservation of approximately 30 percent at 12 to 24 weeks) and pilot variance, the posterior probability of meeting the primary endpoint after observing the interim cohort is:

PP(success)=∫θ​P(final success∣θ,interim data)⋅π(θ∣interim data)dθ

A pre-specified PP threshold below 0.20 at 50 percent enrollment would have stopped the trial closer to 40 randomized than 78, conserving roughly 30 patients of exposure and shaving an estimated 14 months from enrollment burden.

Second, biomarker enrichment by stage and proinsulin-reactive T cell baseline would have rebuilt the case for a true responder population. Pilot data showed C-peptide stability concentrated in HLA-DR4 positive participants with measurable proinsulin-specific CD8+ T cells at baseline, a subgroup making up roughly 40 percent of the pilot cohort. Restricting SUNRISE to that group at an NNS of approximately 2.5 would have raised statistical power for the prespecified effect from approximately 0.40 to 0.75 at the same N. The single highest leverage change would have been to enroll stage 2 (single autoantibody plus dysglycemia) participants rather than new-onset T1D patients within five years of diagnosis.

What this means for similar programs

Antigen-specific tolerance platforms targeting proinsulin, GAD65, or insulin B-chain peptide alone have now failed disease-modification primary endpoints in 4 of 4 randomized Phase 2 trials in established T1D (Diamyd GAD65, Ludvigsson 2012,, oral insulin TrialNet TN07, Krischer 2017,, nasal insulin INIT II, Harrison 2018,, and TOL-3021 SUNRISE). The Phase 2 to Phase 3 conditional success rate for T1D disease-modification programs sits at approximately 14 percent across the last decade (Wong et al., Biostatistics 2019). The realistic path forward likely combines antigen-specific tolerance with broad immunomodulation (anti-CD3 or JAK inhibition) at a defined pre-clinical stage of disease, rather than monotherapy in clinically apparent T1D.

Open questions

How much of the SUNRISE immunological response data will Tolerion publish given the company's size and the program's status? Without the per-protocol CD8+ tetramer and Treg flow data the field is left to infer rather than learn. Does the timing-mismatch hypothesis extend to other peripheral tolerance modalities (microparticles, ImmTOR, peptide-MHC nanoparticles) currently in Phase 1 for T1D? And does the 24-week MMTT C-peptide endpoint remain fit for purpose for slow-onset T1D phenotypes, or should the field accept a longer 12 to 18 month preservation window as primary, accepting larger trials in exchange for more biology?

Sources

1. • ClinicalTrials.gov NCT03895437 (Tolerion SUNRISE record, status update 2026-05-14). https://clinicaltrials.gov/study/NCT03895437 - Open Targets Platform, INS-T1D association (ENSG00000254647, MONDO_0005147), overall score 0.690, accessed 2026-05-20. https://platform.opentargets.org/ - Roep BO et al. Plasmid-encoded proinsulin preserves C-peptide while specifically reducing proinsulin-specific CD8+ T cells in type 1 diabetes. Sci Transl Med 2013, 5(191), 191ra82. https://- Herold KC et al. An anti-CD3 antibody, teplizumab, in relatives at risk for type 1 diabetes. NEJM 2019, 381, 603-613. https://- Culina S et al. Islet-reactive CD8+ T cell frequencies in the pancreas, but not in blood, distinguish type 1 diabetic patients from healthy donors. Sci Immunol 2018, 3(20), eaao4013. https://- Bender C et al. The healthy exocrine pancreas contains preproinsulin-specific CD8+ T cells that attack islets in type 1 diabetes. Nature 2022, 604, 317-323. https://- Ludvigsson J et al. GAD65 antigen therapy in recently diagnosed type 1 diabetes mellitus. NEJM 2012, 366, 433-442. https://- Krischer JP et al. Effect of oral insulin on prevention of diabetes in relatives of patients with type 1 diabetes mellitus (TrialNet TN07). JAMA 2017, 318(19), 1891-1902. https://- Harrison LC et al. Pancreatic beta-cell function and immune responses to insulin after administration of intranasal insulin (INIT II). Diabetes Care 2018, 41(8), 1671-1677. https://- Wong CH et al. Estimation of clinical trial success rates and related parameters. Biostatistics 2019, 20(2), 273-286. https://- Pugliese A et al. The insulin gene is transcribed in the human thymus and transcription levels correlate with allelic variation at the INS VNTR-IDDM2 susceptibility locus. Nat Genet 1997, 15, 293-297. https://- ChEMBL Molecule database, accessed 2026-05-20 (TOL-3021 not indexed as small molecule). https://www.ebi.ac.uk/chembl/ - openFDA FAERS, proinsulin search, accessed 2026-05-20. https://api.fda.gov/drug/event.json.