Ladarixin in recent-onset type 1 diabetes: a subgroup that did not replicate

What was tried

Dompe Farmaceutici ran GLADIATOR (NCT04628481), a Phase 2 randomised, double-blind, placebo-controlled trial of oral ladarixin in people aged 14 to 45 with recent-onset type 1 diabetes and low residual beta cell function at baseline. Ladarixin (ChEMBL189475, DF2156A) is an allosteric non-competitive inhibitor of the interleukin-8 receptors CXCR1 and CXCR2. Participants received 400 mg twice daily for 13 cycles of 14 days on and 14 days off, or matching placebo, randomised 2:1 under quadruple masking. Entry required a first insulin dose within 180 days, at least one diabetes-related autoantibody, fasting C-peptide below 0.205 nmol/L and peak stimulated C-peptide above 0.2 nmol/L. The primary endpoint was change from baseline to month 6 in the 2-hour C-peptide area under the curve during a mixed-meal tolerance test.

The trial started on 12 January 2021 and reached primary completion on 31 March 2025. In total 289 participants entered the run-in and 141 were randomised, 95 to ladarixin and 46 to placebo. ClinicalTrials.gov records the reason for stopping as "The study was stopped due to futility as per protocol", and results were posted on 31 August 2026.

The biological hypothesis

The CXCL8 axis signalling through CXCR1 and CXCR2 recruits neutrophils and other myeloid cells into inflamed tissue. Citro and colleagues reported in 2015 that reparixin and ladarixin prevented inflammation-mediated islet damage in multiple low-dose streptozotocin mice, and both prevented and reversed diabetes in NOD mice, with CXCR2-positive myeloid cells the most reduced leukocyte subpopulation. That work described CXCR1 and CXCR2 as master regulators of diabetes pathogenesis. Alhamar and colleagues added human expression support in 2025, reporting raised IL-8 and IL-8 receptor gene expression in pancreas from donors within two years of diagnosis and higher serum IL-8 and myeloperoxidase in recent-onset patients.

What the hypothesis never had was human genetics. Open Targets scores the CXCR2 association with type 1 diabetes at 0.302, assembled from clinical evidence at 0.484 and literature at 0.264, with no genetic association datatype contributing.

What actually happened

The primary endpoint missed. The adjusted mean difference in change from baseline in log(2-hour C-peptide AUC + 1) at month 6 was -0.133, 95% CI -0.334 to 0.068, p = 0.196, with the point estimate favouring placebo. Every later comparison also crossed zero: month 12 gave 0.021 (95% CI -0.398 to 0.440, p = 0.921), month 18 gave -0.050 (-0.652 to 0.552, p = 0.871) and month 24 gave 0.182 (-8.570 to 8.934, p = 0.967).

Secondary metabolic endpoints tracked the same result. No HbA1c comparison approached significance, the largest being 0.468 percentage points at month 6 favouring placebo (p = 0.528). Mean daily insulin requirement rose in both arms, reaching 0.47 IU/kg/day on ladarixin and 0.46 on placebo at month 24. The proportion needing no insulin was 0 percent on ladarixin at months 12, 18 and 24, against 2.8, 3.2 and 4.0 percent on placebo.

Safety was unremarkable in aggregate. There were no deaths and one serious adverse event in each arm. Non-serious events affected 71 of 94 ladarixin participants and 33 of 46 on placebo, with more infections and gastrointestinal events on drug, including COVID-19 in 25 of 94 against 9 of 46 and dyspepsia in 17 of 94 against 1 of 46. Self-reported severe hypoglycaemia totalled 278 episodes across 94 ladarixin participants and 76 across 46 on placebo, or 2.96 and 1.65 per participant. Those are unadjudicated counts with no statistical comparison attached, so they support no inference about hypoglycaemia risk.

Failure mechanism, best guess

GLADIATOR was a subgroup replication trial, and the subgroup did not replicate. The preceding Phase 2 study (NCT02814838, 76 participants, published by Piemonti and colleagues in 2022) also missed its primary endpoint, reporting a difference of 0.149 nmol/L, 95% CI -0.04 to 0.33, p = 0.122. It then reported a transient metabolic benefit at week 26 in the prespecified subpopulation with fasting C-peptide below the screening median. GLADIATOR selected for exactly that population and found nothing. The two primaries used different scales, raw nmol/L against a log transform, so the estimates cannot be compared numerically, but the direction of the newer and larger one reversed.

Two other explanations remain live and the trial cannot separate them. The intermittent schedule leaves the receptor unblocked for half the treatment period, which is hard to defend against a continuous autoimmune process. Blocking myeloid recruitment may also sit downstream of the autoreactive T cell biology that destroys human beta cells, in which case no schedule would have worked.

How to prevent this next time

The endpoint data support arithmetic rather than modelling, so the levers are design levers.

Treat a subgroup finding inside a trial that missed its primary as a hypothesis, not an effect estimate. The 2022 subgroup was transient and was one of several secondary comparisons in a 76-patient study.

Price the enrichment. Requiring low residual beta cell function reduced 289 run-in entrants to 141 randomised participants, a 51 percent loss before any dosing. The 2:1 allocation left 46 placebo participants and a primary confidence interval half-width of 0.201 log units. Enrichment has to buy more true effect than it costs in precision, and no independent estimate said it would.

Tie the futility rule to the mechanism, not only the endpoint. A target with no human genetic support and an intermittent schedule warrants a pharmacodynamic readout, so a null result can be attributed to the drug or to the hypothesis.

The single highest leverage change would have been requiring an independent replication of the low C-peptide subgroup effect, in a fresh cohort with a prespecified effect size, before committing a 289-patient enrichment trial to it.

What this means for similar programs

The Claidex Mechanistic Risk Score for CXCR2 stands at 43 of 100, band yellow, from phase burden 8.05, archetype severity 9.84, recency 4.25, genetic deficit 10.47 and saturation 10.48. Genetic deficit and saturation dominate, the signature of a heavily drugged receptor with no human genetic association behind it.

Open Targets lists six distinct clinical programmes against CXCR2, including reparixin at Phase 3 and navarixin, SX-682, elubrixin and danirixin at Phase 2. None has reached approval. Within type 1 diabetes the ladarixin franchise has now closed on several fronts: NCT04899271 and NCT05368402 were terminated for low recruitment, NCT05035368 was withdrawn after a design review, and GLADIATOR stopped for futility. Programmes reasoning from myeloid recruitment to beta cell preservation should treat that record as the base rate.

Open questions

Did ladarixin achieve sustained CXCR1 and CXCR2 occupancy on the intermittent schedule, and was any pharmacodynamic marker collected? Does the raised pancreatic IL-8 receptor expression in recent-onset donors drive beta cell loss or follow it? Will the full GLADIATOR publication report the subgroup analysis that motivated the design?

Sources

1. ClinicalTrials.gov. NCT04628481 (GLADIATOR), posted results. https://clinicaltrials.gov/study/NCT04628481.

2. Piemonti L, et al. Diabetes Obes Metab. 2022.

3. Citro A, et al. Diabetes. 2015.

4. Alhamar G, et al. Diabetes Obes Metab. 2025.

5. Open Targets Platform, CXCR2 and type 1 diabetes, accessed 1 September 2026. https://platform.opentargets.org/evidence/ENSG00000180871/MONDO_0005147.

6. ChEMBL. Ladarixin, CHEMBL189475. https://www.ebi.ac.uk/chembl/compound_report_card/CHEMBL189475/.

7. ClinicalTrials.gov. NCT02814838, NCT04899271, NCT05368402, NCT05035368. Available from: https://clinicaltrials.gov/study/NCT02814838.