Oral insulin immunotherapy in children at risk for type 1 diabetes in a randomised controlled trial

Aims/hypothesis Oral administration of antigen can induce immunological tolerance. Insulin is a key autoantigen in childhood type 1 diabetes. Here, oral insulin was given as antigen-specific immunotherapy before the onset of autoimmunity in children from age 6 months to assess its safety and immune response actions on immunity and the gut microbiome. Methods A phase I/II randomised controlled trial was performed in a single clinical study centre in Germany. Participants were 44 islet autoantibody-negative children aged 6 months to 2.99 years who had a first-degree relative with type 1 diabetes and a susceptible HLA DR4-DQ8-containing genotype. Children were randomised 1:1 to daily oral insulin (7.5 mg with dose escalation to 67.5 mg) or placebo for 12 months using a web-based computer system. The primary outcome was immune efficacy pre-specified as induction of antibody or T cell responses to insulin and measured in a central treatment-blinded laboratory. Results Randomisation was performed in 44 children. One child in the placebo group was withdrawn after the first study visit and data from 22 insulin-treated and 21 placebo-treated children were analysed. Oral insulin was well tolerated with no changes in metabolic variables. Immune responses to insulin were observed in children who received both insulin (54.5%) and placebo (66.7%), and the trial did not demonstrate an effect on its primary outcome (p = 0.54). In exploratory analyses, there was preliminary evidence that the immune response and gut microbiome were modified by the INS genotype Among children with the type 1 diabetes-susceptible INS genotype (n = 22), antibody responses to insulin were more frequent in insulin-treated (72.7%) as compared with placebo-treated children (18.2%; p = 0.03). T cell responses to insulin were modified by treatment-independent inflammatory episodes. Conclusions/interpretation The study demonstrated that oral insulin immunotherapy in young genetically at-risk children was safe, but was not associated with an immune response as predefined in the trial primary outcome. Exploratory analyses suggested that antibody responses to oral insulin may occur in children with a susceptible INS genotype, and that inflammatory episodes may promote the activation of insulin-responsive T cells. Trial registration Clinicaltrials.gov NCT02547519 Funding The main funding source was the German Center for Diabetes Research (DZD e.V.) Graphical abstract Supplementary Information The online version contains peer reviewed but unedited supplementary material available at 10.1007/s00125-020-05376-1.


Randomisation and masking
A computer-generated randomisation list was prepared with an allocation ratio of 1:1 (placebo to oral insulin) using a web-based system (https://wwwapp.ibe.med.uni-muenchen.de/randoulette/). All investigators and participants were masked to the treatment allocation. Unblinding was not necessary during the study.

Investigational Medicinal Product
Insulin crystals were provided by Lilly Pharmaceuticals (Indianapolis, Indiana). The investigational medicinal products (insulin and placebo) were manufactured as identical capsules containing either insulin crystals (7.5 mg, 22.5 mg, or 67.5 mg) in microcrystalline cellulose (total capsule content 200 mg) or 200 mg microcrystalline cellulose placebo by InPhaSol, Apotheke des Universitätsklinikums Heidelberg, Germany. The drug packages were sequentially numbered according to the randomly allocated treatment. Parents were instructed to sprinkle the contents of one capsule onto one teaspoon of food (e.g. yogurt, breast milk, or commercial baby food) for administration once daily.

Measurements of immune responses to insulin
Insulin autoantibody (IAA) levels were measured using a competitive radio binding assay [6,7]. A positive response was defined as a value of ≥1.5 and a ≥2-fold increase from baseline. Serum IgG binding to insulin was measured by a non-competitive radio binding assay with protein-G capture of IgG [8]. A positive response was defined as an increase of >10 counts/min from the baseline value. Salivary IgA binding to insulin was measured using a radio binding assay as previously described [9]. A positive response was defined as an increase of ≥ 3-fold from baseline. Serum IgE against insulin was measured using a radio binding assay [10].
CD4 + T cell antigen responses were measured using stored frozen peripheral blood mononuclear cells (PBMCs).
Responses were measured using a dye (Cell Proliferation Dye eFluor 670, eBioscience, San Diego, CAL, USA) dilution assay, quantifying proliferation (eFluor670dim cells) and activation (CD25 + ) after 5 days of culture without or with the antigen insulin that was identical to the insulin administered to the children (50 µg/ml, Lilly Pharmaceuticals) as previously described [9] (ESM Fig. 1). The assay included a median of 12 wells containing 200,000 eFlour670 dye-labelled cells in medium plus insulin and 6 wells with cells and medium alone. The SI was calculated as the number of CD4 + eFluor670dimCD25 + cells per 50,000 acquired live CD4 + T cells in all wells containing insulin relative to the number of CD4 + eFluor670dimCD25 + cells per 50,000 acquired live CD4 + T cells in all wells containing medium alone. A positive sample was defined as an SI of >3. A positive T cell outcome was defined as a positive sample and an increase in SI of >2-fold at any follow-up visit relative to the 5 baseline value. CD8 + T cell proliferation responses to insulin were also measured in the same assay by gating on CD8 + CD4 − T cells (ESM Fig. 1).

Single-cell gene expression profiling of CD4 + T cells responding to insulin
CD4 + T cells that had proliferated, as determined by eFluor® 670 dilution, and displayed CD25 upregulation were identified as responding cells and were single-cell-sorted directly into 96-well microplates containing 5 µl of PBS prepared with diethylpyrocarbonate-treated water. For samples that had an SI against insulin above 3, cells were processed for gene expression. cDNA was synthesized directly from cells using qScript™ cDNA Supermix (Quanta Biosciences, Gaithersburg, MD, USA). Total cDNA was pre-amplified for 18 cycles, with 1 cycle of denaturation at 95 °C for 1 min, followed by cycling at 95 °C for 15 s, 60 °C for 1 min, and 72 °C for 1.5 min, followed by one cycle of 72 °C for 7 min, with TATAA GrandMaster Mix (TATAA Biocenter, Göteborg, Sweden) in the presence of 76 primer pairs at a final volume of 35 µl (ESM Table 7 Low ROX (Bio-Rad) and 5 µM of primers in each assay. The primers and target genes are listed in Table S7.
Raw data were analyzed using Fluidigm Real-Time PCR analysis software and GenEx Pro 5.3.6 Software (MultiD, Göteborg, Sweden). Additional data analysis was done using KNIME 2.5.2 software (KNIME AG, Zürich, Switzerland). Analysis of multivariate gene expression patterns was performed by Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP) [11] and unsupervised WARD hierarchical clustering (hclust) on the pre-processed Ct values. For pre-processing, a linear model was used to correct for 6 potential confounding effects, which can mask relevant biological variability [12]. In brief, batch effects (dummy coding for each plate/batch) were modelled jointly with dose effects by regressing out the effect of plates on each individual gene while controlling for dose in order to obtain a corrected gene expression dataset.

Plasma inflammatory markers
Inflammation-related protein biomarkers were determined after unblinding of participants. Measurements were performed by proximity extension assay using the Olink inflammation panel (Olink, Uppsala, Sweden) following manufacturer's instructions.

Stool microbiome
Stool samples were collected at home at the day of the visit or within two days before or after the visit at baseline, 6 months, and 12 months, into tubes containing ethanol that were provided to the parents. The samples were brought to the visit, aliquoted, and stored at −80 °C. Alternatively, samples could be sent to the central laboratory with guaranteed delivery within 24 h. The bacterial component of the microbiome in each stool sample was analyzed by 16S rRNA gene compositional analysis, as previously described [13]. To generate 16S rDNA data, genomic bacterial DNA was extracted from the samples using MO BIO PowerMag Soil DNA Isolation Kit (Qiagen, Venlo, Netherlands). The 16S rDNA V4 region was amplified by PCR and sequenced in the MiSeq platform (Illumina, San Diego, CA, USA) using the 2 × 250 bp paired-end protocol. The primers used for amplification contained adapters for MiSeq sequencing and dual-index barcodes so that the PCR products may be pooled and sequenced directly [14], targeting at least 10,000 reads per sample. The standard pipeline for processing and analysing the 16S rDNA gene data incorporated phylogenetic and alignment-based approaches to maximize data resolution. The read pairs were demultiplexed based on the unique molecular barcodes, and reads were merged using USEARCH v7.0.1001 [15]. 16S rRNA gene sequences were assigned into Operational Taxonomic Units (OTUs) or phylotypes at a similarity cut off value of 97% using the UPARSE pipeline.
Abundances were recovered by mapping the demultiplexed reads to the UPARSE OTUs.
A subset of samples with sufficient material was selected for metagenomic whole genome shotgun (WGS) sequencing for deeper characterisation. Metagenomic WGS sequencing utilized the same extracted bacterial genomic DNA used for 16S rDNA compositional analysis. For WGS, individual libraries constructed from each 7 sample were loaded into the HiSeq platform (Illumina, San Diego, CA, USA) and sequenced using the 2x100 bp pair-end read protocol. The process of quality filtering, trimming and demultiplexing was carried out by in-house pipeline developed by assembling a number of publicly available tools such as Casava v1. 8

.3 (Illumina, San
Diego, CA, USA) for the generation of fastqs, Trim Galore and cut adapt for adapter and quality trimming, and PRINSEQ for sample demultiplexing. In addition, Bowtie2 v2.2.1 [16] was used to map reads to custom databases for bacteria, viruses, human, and vectors. Reads whose highest identity match was not bacterial were removed from subsequent analysis. For bacterial reads, the highest identity match was chosen. If there were multiple top hits, the lowest common ancestor was determined.

Recording of adverse events
Throughout the study, the investigators recorded any adverse events using an adverse event clinical report form, regardless of the event's severity or relation to the study drug or study procedure. The families were instructed to note any symptoms of hypoglycaemic events such as trembling, sweating or impaired consciousness after study drug intake. Hypoglycaemia was defined as a blood glucose level <2.78 mmol/l (<50 mg/dl).

Statistical comparisons
Additional analyses were planned to compare the immunological outcomes in children with the INS AA genotype and treatment effects on the stool microbiome. These and all other analyses were considered exploratory. An interaction between INS genotype and treatment on immunological responses to insulin was assessed using the Cox proportional hazards model. All analyses comparing responses in relation to monocyte CD169 expression, and analyses of cell frequency and plasma inflammatory markers were defined post-hoc. Spearman's correlation was calculated to assess the correlation between two continuous variables. Differences between groups' centroids defined by a principal component analysis (PCA) were assessed using a permutational multivariate analysis of variance (PERMANOVA). Analysis of age relationships to cell population frequencies included a linear mixed model with the cell frequencies as fixed effects and the children identification numbers as a random effect was fitted to predict the age. Stool analyses were conducted to characterize differences in the microbiome between the two treatment groups, including stratification by INS AA genotype and to determine the relationship of the microbiome to the immune responses in blood. Differences in beta diversity were visualized using a principal coordinates analysis (PCoA) followed by a PERMANOVA to assess differences between groups.

IgE
IgE concentrations were above the reference limits in 3 children at the 12 months visit; all three children were in the placebo group. Children in the group receiving oral insulin showed no change in measured IgE (Median difference = 0.00). No child had IgE to insulin.

Laboratory analysis
No significant changes in blood cell counts were observed. Monocyte count at baseline was the only parameter with values that were significantly different between the children in the placebo group and the oral insulin group (median 7% versus 5%; p = 0.0089, ESM Table 3a). No differences in blood chemistry values were observed except for GGT at 12 months with higher values in children in the placebo group (median 11 U/l vs 9 U/l, p = 0.0063, ESM Table 3b).  Table 4). The severity of adverse events was similar between the two groups. There were six serious adverse events, four in the oral insulin group and two in the placebo group, none of which were considered related to the study drug. By system organ class, the frequency of skin and subcutaneous tissue disorders was greater in the oral insulin group (12 events in 8 children) than in the placebo group (1 event in 1 child; p = 0.01; ESM Fig. 2). These included diaper rash, erythema, eczema, pruritus, and urticaria (ESM Table 4). The overall frequency of skin and subcutaneous tissue disorders among all reported adverse events was 4.3% and all of these adverse events were classified as mild (grade 1) and resolved.

Protocol violations
There were 332 protocol deviations, all except one classified as minor. Most protocol violations were associated with missing values regarding single blood count parameters, missing parameters during blood glucose monitoring, missing parameters during the physical examination or an exceeded time window for study visits.
The one protocol violation judged as major was the dispensation of a wrong medication package to one subject.
However, after unblinding it turned out that this wrongly dispensed medication package was from the same medication group (placebo) the child had received in the trial.      Severe and undesirable 2 1 (4.