This study was a phase 2, randomised, placebo-controlled, double-blinded trial conducted at the Steno Diabetes Center Copenhagen, Denmark. The trial consisted of a screening period of up to 6 weeks following diagnosis and a treatment period of 52 weeks. Primary and secondary outcomes were assessed at baseline and at 4, 12, 26 and 52 weeks after the initiation of trial treatment.

Before enrolling any participants, the trial protocol was approved by the local ethics committee and the Danish Medicines Agency under the EudraCT number: 2019-004434-41. The trial was conducted in accordance with the principles of the Declaration of Helsinki and Good Clinical Practice. An independent data monitoring committee oversaw the trial.

Participants were recruited from Steno Diabetes Center Copenhagen and Bispebjerg Hospital (and other outpatient clinics located in the Capital Region of Denmark) between June 2020 and April 2022. To obtain representation of the broader population of interest both male and female participants were included in the study. Sex was self-reported and cross-referenced with information in the participant’s medical record. Ethnicity, religion and socioeconomic status were not considered as factors of eligibility. However, the recruiting centres are known to serve a diverse patient population in terms of these factors.

All participants were orally informed about the trial and provided with written information before giving written informed consent. Eligible participants were aged 16–40 years, diagnosed with stage 3 type 1 diabetes according to the ADA criteria, and able to undergo randomisation within 6 weeks from the first insulin injection. Other inclusion criteria were the presence of at least one positive diabetes-specific autoantibody, no contraindications to receiving fenofibrate, and a peak C-peptide level >0.2 nmol/l after a 2 h mixed-meal tolerance test (MMTT) (a complete overview of inclusion criteria is provided in electronic supplementary material [ESM] Methods).

In a block design with a block size of 4, 58 participants were randomly assigned in a 1:1 ratio to receive either fenofibrate or placebo. Pre-packed pill containers with identical tablets of fenofibrate or placebo were supplied by the Pharmacy of the Capital Region of Denmark. The randomisation sequence was computer-generated, with fixed numbers of treatments in each block, to balance the treatment allocation over time. Blinding was maintained for participants, their healthcare providers, and all staff involved in handling outcome samples and assessment. The allocation codes were kept in sealed envelopes and remained concealed until the unblinding phase. Fenofibrate was administered as a daily oral dose of 160 mg, based on its pharmacokinetics, pharmacodynamics and safety data, and in alignment with a case report showing preserved beta cell function 2 years after diagnosis in a patient with type 1 diabetes on daily treatment with 160 mg fenofibrate [8].

Participants received individually numbered pill containers at baseline and at 4, 12 and 26 weeks after randomisation. Compliance was assessed by counting the number of returned pills at each trial visit after randomisation.

Trial visits were conducted in the morning following an overnight fast. Long-acting insulin was administered in the evening prior to the visit, and no rapid-acting insulin was administered within 2 hours preceding the start of the visit.

An MMTT was conducted at all visits to assess beta cell function. Participants ingested a standardised liquid meal, and venous blood samples for C-peptide and glucose analyses were collected at 0, 15, 30, 60, 90 and 120 min after the start of ingestion.

Glucose variability was evaluated using a continuous glucose monitoring (CGM) system of the FreeStyle Libre type, which participants wore from baseline to the 52 week follow-up visit. Compliance with the CGM device was assessed at each visit. If the sensor had been active for less than 70% of the time, the data were considered as missing data.

The use of exogenous insulin was self-reported by the participants for 7 days leading up to each visit using the FreeStyle Libre app on their mobile device.

The primary outcome was the change in endogenous insulin production from baseline to after 52 weeks of treatment, assessed by AUC values for C-peptide levels following an MMTT [9].

Predefined secondary outcomes were changes in peak C-peptide following an MMTT; HbA1c; glucose outcomes measured with CGM (percentage of time spent with glucose concentrations <3.0, 3.0–3.8, 3.9–10.0, 10.1–13.9 and >13.9 mmol/l, and number of hypoglycaemic events [glucose level <3.9 mmol/l persisting for more than 15 min] [10]); daily insulin dose; percentage of participants in remission at week 52 (HbA1c <53 mmol/mol or 7.0% and an insulin use ≤ 0.4 U kg-1 day-1) [11]; and beta cell stress assessed by total proinsulin/C-peptide (PI/C) ratio [12]. Proinsulin levels were assessed using a validated ELISA assay. Due to high cross-reactivity between intermediates and intact proinsulin, total proinsulin levels were used for PI/C ratio [cross-reactivity with intact proinsulin of 74% for split (32–33), 65% for des (31, 32), 78% for split (65–66), and 99% for des (64, 65)-proinsulin].

Autoantibodies against GAD65, IA-2, IAA and ZnT8 were detected using the ‘Antibody detection by agglutination-PCR’ method (see ESM Methods). Safety outcomes were assessed by adverse events during the 52 weeks of intervention. At each visit after randomisation, participants were asked about adverse events in general, and specific adverse events known to be associated with fenofibrate.

For lipidomics, plasma samples were combined with NaCl and chloroform/methanol (2:1), after which lipid-containing chloroform was analysed. Data were processed using the open-source software Skyline (see ESM Methods).

Human pancreatic islets were obtained from a population of seven non-diabetic organ donors (see ESM Human Islets Checklist). From these islets, six subpopulations were exposed to proinflammatory cytokines (50 U/ml recombinant human IL-1β and 1000 U/ml recombinant human IFN-γ) with or without fenofibrate before evaluating cell death and three subpopulations were used for glucose-stimulated insulin secretion (GSIS). One subpopulation was lost during the cell death experiment. Cell death was quantified by measuring cytoplasmic histone-associated DNA fragments (see ESM Methods).

We based our sample size on an ANOVA design to detect a between-group difference in 2 h C-peptide AUC after 52 weeks of treatment with either fenofibrate or placebo, with a 1:1 allocation. Our estimates followed the TrialNet Study Group guidelines for ANOVA models in clinical trials involving newly diagnosed type 1 diabetes [13]. Due to the expected skewness in the distribution of C-peptide AUC values, we performed a logarithmic transformation (log[x+1]) to normalise the data. We assumed a geometric-like mean of 0.281 nmol/l in the placebo group at week 52, with variability similar to that reported by the TrialNet Study Group [13]. For a two-tailed test to detect a 50% between-group difference at a 5% significance level and 85% power, our estimates indicated a required sample size of 58 participants, accounting for a dropout rate of no more than 10%.

Data from all participants who received trial treatment and had at least one analysable value available were included in the statistical analyses for both primary and secondary outcomes.

For the primary outcome, we applied a mixed model for repeated measures fitted on log[x+1]-transformed C-peptide AUC data from each visit, with the back-transformed mean value reported as the geometric-like mean (using exp(y) −1). The response variable was post-baseline log[AUC+1] with sex, time and treatment group as fixed effects, participant as random effect, and age at baseline, time to randomisation (defined as the time from first recognised symptoms of hyperglycaemia to the baseline visit) and C-peptide level at baseline as continuous covariates. For participants with incomplete data, missing data were assumed to be missing at random. Missing data were accounted for by using the mixed model for repeated measures.

A similar mixed model approach was used to assess secondary outcomes. For binary dependent variables, a logistic regression model was employed. The model was adjusted for treatment group, sex, age at baseline, and time to randomisation to assess odds ratios (ORs) and 95% CIs. For human pancreatic islet analysis, data were log[x+1]-transformed, and effects were estimated using a mixed linear model.

All reported p values are two-sided with a statistical significance level at 5%. Multiple comparisons were adjusted using the Benjamini–Hochberg approach. Statistical analyses were conducted using SPSS software (IBM SPSS Statistics version 25.0, IBM, Armonk, NY, US).