Siplizumab in new-onset type 1 diabetes: a dose-finding study that selected on the wrong half of the pharmacology

What was tried

The Immune Tolerance Network ran DESIGNATE (ITN095AI, NCT05574335) under National Institute of Allergy and Infectious Diseases sponsorship, an open-label randomised dose-finding study of siplizumab in recently diagnosed type 1 diabetes. Siplizumab (ChEMBL CHEMBL2108742, also recorded as MEDI-507 and TCD601) is a humanised monoclonal antibody against CD2. Eligible participants were 8 to 45 years old, diagnosed within 550 days, positive for at least one islet autoantibody, and had a peak stimulated C-peptide above 0.15 nmol/L. They were randomised 1:1:1:1 to weekly subcutaneous siplizumab at 0.08, 0.12, 0.18 or 0.22 mg/kg for 12 weeks, with a paediatric cohort planned.

The design inverted the usual order of dose selection. Rather than choosing a dose on tolerability or exposure, DESIGNATE chose on an immunological readout: a participant counted as a responder if, between week 0 and week 12, the blood CD4 regulatory T cell to effector memory T cell ratio rose by at least 75 percent and PD1+TIGIT+ frequency within circulating CD4 effector memory T cells rose by at least 20 percent. Stimulated 2-hour C-peptide AUC and daily insulin use were secondary. The study started on 18 May 2023, activated 20 US sites, and closed with a primary completion date of 14 May 2025.

The biological hypothesis

CD2 (Ensembl ENSG00000116824) is a costimulatory glycoprotein on T and NK cells that binds CD58 and helps organise the immunological synapse. Expression is higher on memory T cells than on naive cells, and that difference is the therapeutic premise: an antibody engaging CD2 should strip effector memory T cells while sparing naive cells and resting regulatory T cells (doi:10.3389/fimmu.2020.01090). In allogeneic mixed lymphocyte reactions, siplizumab preferentially reduced CD4 and CD8 effector memory T cells and enriched proliferating FoxP3-high cells, with TCR sequencing showing selective expansion of donor-reactive regulatory T cells (doi:10.1111/ajt.15533).

The clinical precedent was alefacept, an LFA-3 fusion protein that also engages CD2 and remains the only approved CD2-directed agent in Open Targets. In T1DAL, 49 people with new-onset disease received alefacept or placebo. The 12-month primary endpoint, change in mean 2-hour C-peptide AUC, was not met (increase of 0.015 nmol/L versus a decrease of 0.115 nmol/L on placebo, p=0.065), while the 4-hour AUC, insulin use and hypoglycaemia rate favoured alefacept (doi:10.1016/S2213-8587(13)70111-6). At 24 months the 4-hour AUC fell by 0.134 nmol/L versus 0.368 nmol/L (p=0.002), with memory T cells depleted and regulatory T cells preserved (doi:10.1172/JCI81722). Alefacept was later withdrawn commercially, so DESIGNATE tried to recover a mechanism that had produced signal but never a registered indication.

Genetic support for this pairing was thin. The Open Targets association score for CD2 and type 1 diabetes was 0.10166, from literature (0.102) and clinical (0.162) evidence with no genetic association contribution. CD2 scores far higher against hypothyroidism (0.451), rheumatoid arthritis (0.431) and Graves disease (0.429), where genetic association carries most of the weight.

What actually happened

Seventeen adults were screened and 8 were randomised across 11 of the activated sites. One never received a dose, leaving 7 treated. No children were enrolled before the study stopped, so the paediatric cohort never opened. Six participants had viable samples after two consecutive doses and were evaluable for the primary endpoint.

Three of those 6 met the composite signature. The PD1+TIGIT+ criterion was met by every evaluable participant in every arm. The regulatory T cell ratio criterion was met at 0.08, 0.12 and 0.18 mg/kg but not in the single evaluable participant at 0.22 mg/kg.

Serious safety reporting was clean, with no serious adverse events and no deaths in any arm. The reported burden was laboratory: decreased CD4 lymphocytes in 7 of 7 treated participants, decreased lymphocyte count in 6 of 7, decreased white blood cell count in 4 of 7, and decreased neutrophil count in 2 of 7. Infections were COVID-19 in 2 participants and single cases of infectious mononucleosis, pharyngitis and urinary tract infection.

The registry states the stopping sequence plainly. The study was on an enrolment hold for greater than anticipated lymphodepletion, and the pharmaceutical partner then declined to reopen its own type 1 diabetes trial and ceased development of siplizumab in autoimmunity. No summary results were posted for the secondary metabolic endpoints, so nothing can be said about C-peptide or insulin use.

Failure mechanism, best guess

The archetype is translational mismatch. The programme assumed that the dose producing the desired regulatory shift would also produce a tolerable depth of depletion, and those properties came apart in humans at the doses tested. The selection endpoint was a ratio, and a ratio improves when the denominator falls. The denominator here was effector memory T cells, the same population whose loss defines lymphodepletion. Every treated participant recorded decreased CD4 lymphocytes, so the signature and the safety-limiting effect were driven by the same pharmacology.

The preclinical package made this hard to see in advance. It was dominated by in vitro mixed lymphocyte reactions, which report proportions within a culture and cannot report an absolute lymphocyte count in a person. Siplizumab also depletes through antibody-dependent cellular cytotoxicity and drives NK cell fratricide (doi:10.3389/fimmu.2021.599526), an effector mechanism whose intensity depends on Fc receptor biology that those assays do not reproduce. The partner exit followed the hold rather than causing it.

How to prevent this next time

The posted data do not support a quantitative model: there is no endpoint-level metabolic result, the evaluable denominator is 6, and no placebo arm exists. The available levers are design and governance levers.

Pair every ratio endpoint with an absolute floor. Both DESIGNATE criteria were proportional and both could be satisfied by depletion alone. The rule should have carried a hard stop on absolute CD4 count, so that no dose could qualify while breaching the depletion boundary.

Enrich on the mechanism, not the diagnosis. CD2 carries no genetic association evidence for type 1 diabetes in Open Targets. A baseline CD2-high effector memory fraction is a candidate entry criterion. Separately, an academic trial whose supply sits with one commercial partner inherits that partner's portfolio risk, which belongs in feasibility planning.

The single highest leverage change would have been writing the dose-selection rule as a joint constraint on the regulatory ratio and an absolute lymphocyte floor, so that greater than anticipated lymphodepletion registered as a failed dose rather than as a successful signature.

What this means for similar programs

CD2 remains sparsely populated. Open Targets records 2 clinical programmes, alefacept at approval and siplizumab at Phase 2, which is why the saturation component of this target's Mechanism Risk Score is low at 4.95 and the total sits at 36, in the yellow band. The mechanism is not refuted. What this trial removes is the assumption that a proportional immunophenotype can gate dose selection for a depleting antibody. Programmes using any Fc-competent depleting antibody in autoimmunity should expect the regulatory-cell enrichment they want to be partly an artefact of the depletion they are trying to bound.

Open questions

Did the 3 participants meeting the signature differ in absolute CD4 count from the 3 who did not, and was the enrolment hold triggered by a breach in one arm or by an aggregate trend? Neither is recoverable from the registry.

Were any secondary metabolic data collected before the hold, and will they be published? Seven participants cannot support an efficacy claim, but paired immunophenotype and C-peptide trajectories would inform the next CD2 programme.

Sources

1. • ClinicalTrials.gov, NCT05574335, DESIGNATE (ITN095AI), full record and posted results section. https://clinicaltrials.gov/study/NCT05574335 (accessed 2026-07-29) - Open Targets Platform, CD2 (ENSG00000116824), target record and association with type 1 diabetes mellitus (MONDO_0005147), score 0.10166. https://platform.opentargets.org/target/ENSG00000116824 (accessed 2026-07-29) - ChEMBL, siplizumab, CHEMBL2108742, antibody, maximum phase 2. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL2108742/ (accessed 2026-07-29) - openFDA drug event API, medicinal product SIPLIZUMAB, 4 total reports, 4 serious. https://api.fda.gov/drug/event.json (accessed 2026-07-29) - Binder C, et al. CD2 Immunobiology. Front Immunol.

2. - Podestà MA, et al. Siplizumab selectively depletes effector memory T cells and promotes a relative expansion of alloreactive regulatory T cells in vitro. Am J Transplant.

3. - Binder C, et al. Siplizumab, an anti-CD2 monoclonal antibody, induces a unique set of immune modulatory effects compared to alemtuzumab and rabbit anti-thymocyte globulin in vitro. Front Immunol.

4. - Binder C, et al. Siplizumab induces NK cell fratricide through antibody-dependent cell-mediated cytotoxicity. Front Immunol.

5. - Rigby MR, et al. Targeting of memory T cells with alefacept in new-onset type 1 diabetes (T1DAL study): 12 month results of a randomised, double-blind, placebo-controlled phase 2 trial. Lancet Diabetes Endocrinol.

6. - Rigby MR, et al. Alefacept provides sustained clinical and immunological effects in new-onset type 1 diabetes patients. J Clin Invest.

7. .