Type 1 diabetes is caused by islet autoimmunity and declining insulin secretion. Environmental factors contribute to the condition1 in addition to genetic predisposition.2 Insulin depletion typically develops more rapidly in children than in adults, but the rate of progression to insulin deficiency varies greatly.

The loss of beta cell function varies with age,3 sex,3 genotype,4 body mass index (BMI),5 Haemoglobin A1c (HbA1c),6 as well as ketoacidosis and islet cell autoantibodies at diagnosis. A residual beta cell function carries great patient benefit as even a small preserved insulin secretion is associated with fewer episodes of ketoacidosis7 and serious hypoglycemia and less risk of late diabetes complications.8 We have recently confirmed the association between low age, ketoacidosis, higher HbA1c, high titers of glutamic acid decarboxylase antibody (GADA) and islet antigen-2 antibody (IA-2A), and a more rapid loss of endogenous insulin secretion.9 Moreover, we found that a rapid decline in C-peptide is associated with an increased incidence of severe hypoglycemia. Conversely, high BMI SDS (SDS=Standard deviation score, deviation of BMI from a reference population), low HbA1c during the first years, and higher frequency of the HLADR3 genotype were associated with long-term preservation of C-peptide.9 We have also previously found that patients with long-standing type 1 diabetes and preservation of C-peptide have increased levels of circulating interleukin (IL)-35 and a higher frequency of IL-35+ regulatory T cells (Tregs), suggestive of an altered immunological phenotype.10

Identifying patterns of biomarkers or individual markers that can predict disease progression rate at diagnosis could reveal novel insights regarding the pathogenesis of type 1 diabetes. Furthermore, predicting the progression rate at type 1 diabetes onset holds great value in appropriately randomizing patients to clinical intervention trials aiming to preserve beta cell function and, ideally, selecting the most appropriate beta cell preservation therapy. Therefore, we have investigated whether multiplexing of biomarkers at diagnosis can predict the rate of loss of C-peptide secretion in a cohort of children with new-onset type 1 diabetes.

The study is based on a retrospective, observational study including 46 children (born between 1989 and 2007) with newly diagnosed type 1 diabetes (year of diagnosis between 2004 and 2017) who initiated insulin treatment on admission. The participants were regularly followed at the Crown Princess Victoria Children’s Hospital in Linköping, Sweden. At the age of 18 years, participants were transferred to the diabetes clinic for adults. The diagnosis of type 1 diabetes was based on the criteria set by the American Diabetes Association for diagnosis and classification. At baseline, the T1D group was also compared with a sex-matched healthy control group of similar age, Healthy (N=45), from the ABIS (All Babies in Southeast Sweden) cohort.

The hypothesis was that multiplexed biomarkers can differentiate patients with rapid loss of C-peptide secretion from those with a slower decline, indicating distinct biological pathways or disease mechanisms associated with the progression of type 1 diabetes.

The aim was to investigate biomarker differences in relation to the course of residual C-peptide and whether or not multiplexed biomarkers could distinguish patients with a rapid loss of C-peptide secretion. Based on longitudinal stimulated C-peptide data, the cohort was divided into two groups, Rapid (n=20) and Slow (n=26). Rapid progression was defined as a loss of C-peptide secretion within 30 months following type 1 diabetes debut (cut-off <0.03 nmol/L). The two type 1 diabetes groups, Rapid and Slow, were also compared over time.

Descriptive data of age, sex, HbA1c, blood glucose, blood pH, and C-peptide levels at the time of diagnosis before initiating insulin treatment were collected from electronic medical records. During follow-up visits (10 days, 1, 3, 9, 18, 24, and 30 months, and 3, 4, 5, and 6 years after diagnosis), additional data were recorded, including weight, height, HbA1c, and insulin dosage (expressed as units per kilogram of body weight per 24 hours). In addition, BMI and BMI SDS, adjusted for age and sex, were automatically generated using the SWEDIABKIDS register.11

Mixed meal tolerance tests (MMTTs) were performed under fasting conditions in the morning. Baseline measurements of C-peptide and glucose were obtained, followed by sampling at 30-minute intervals during the 120-minute test. Short-acting insulin administration was withheld for at least 6 hours before the MMTT. The composition of the mixed meal changed over time, transitioning from a standardized breakfast to a standardized liquid meal based on the participant’s body weight. C-peptide concentrations were measured using a time-resolved fluoroimmunoassay with a lower detection limit of 0.03 nmol/L, and undetectable C-peptide levels were assigned a 0.01 nmol/L value for statistical analysis.

HbA1c and blood glucose measurements were performed at the Department of Clinical Chemistry, Linköping. The laboratory is certified by Swedac, a Swedish government authority. As of October 2010, HbA1c is analyzed using the International Federation of Clinical Chemistry and Laboratory Medicine reference method and expressed in mmol/mol. Before October 2010, analyses were based on the Mono S standard and expressed in percentage. HbA1c analyses performed with the Mono S standard were recalculated using the following expression: (International Federation of Clinical Chemistry and Laboratory Medicine (IFCC); mmol/mol) = 10.45 × HbA1c (Mono S; %)−10.62 (https://ngsp.org/convert1.asp).

Autoantibodies GADA (detection limit 5 IU/mL) and IA-2A (detection limit 7.5 kU/L) were analyzed using two-sided ELISA test kits from RSR (RSR, Cardiff, UK) in serum according to the instructions from the manufacturer. Samples negative for ELISA IA-2A were further analyzed with a high-sensitivity IA-2A radio binding assay. Recombinantglutamic acid decarboxylase 65 (GAD65) and islet antigen 2 (IA-2) were labeled with [35S] methionine (GE Healthcare Life Sciences, Amersham, UK) by in vitro-coupled transcription and translation in the TNT SP6 coupled reticulocyte lysate system (Promega, Southampton, UK) as described. Full-length cDNA coding for human GAD65 in the pTNT vector (Promega) (pThGAD65) or the intracellular domain (aa 603–980) of IA-2 in the pSP64 Poly(A) vector (Promega) (IA-2ic) was used. These analyses were conducted at the Department of Clinical Chemistry, Skåne University Hospital, Malmö, Sweden. The intra-assay coefficient of variation for duplicates in the GADA assay was 7% and in the IA-2A 11%. In the Diabetes Autoantibody Standardization Program 2010 workshop, our laboratory was among the top-ranking laboratories for GADA in workshop sensitivity (80%) and specificity (99%) and the top-ranking laboratory for IA-2A in workshop sensitivity (60%) and specificity (99%).12

Protein analysis was performed in undiluted EDTA plasma by multiplex PEA at Olink Proteomics AB (Uppsala, Sweden) according to the manufacturer’s protocol. Two validated 92-plex panels, Olink IMMUNE ONCOLOGY (IMO) and Olink IMMUNE RESPONSE (IRE, online supplemental tables 1 and 2), were used to measure markers associated with inflammation and active immune and cytokine response. The IMO panel consists of proteins involved in tumor immunity, chemotaxis, tissue remodeling, apoptosis, cytotoxicity, metabolism, and autophagy. The IRE panel is focused on key proteins involved in adaptive immune response, viral defense, inflammation, and cytokine signaling. These panels were chosen based on their complementary immune profiles. Samples were collected and analyzed in patients with type 1 diabetes at 10 days, 3, 9, and 18 months after diagnosis. Additional samples were collected at 30, 48, and 72 months in those with remaining C-peptide at 30 months (Slow). Only baseline samples were analyzed in healthy controls.

Statistical analyses were performed at Olink using R V.4.1.2 (2021-11-01), apart from baseline characteristics that were analyzed in-house using RStudio V.2022.12.0+353. Values are given as means±2 SDs. P values <0.05 after correction for multiple testing with the Benjamini and Hochberg procedure (unless otherwise stated) were considered statistically significant. Samples were analyzed in three different batches. A principal component analysis (PCA) plot was used to identify outliers among 10 bridging samples to ensure accuracy. Based on this quality control (QC) plot, the IMO panel had no outliers, while one baseline sample in the IRE panel deviated from the remaining samples. This sample was hence removed from the analysis of the IRE panel. All samples were included in longitudinal analyses. The samples were normalized between batches using the median difference as an adjustment factor for the limit of detection. The normalization process was evaluated using density plots and was observed to make the sample distributions more similar between batches.

PCA was further used to identify the protein patterns of each group. A Welch two-sample t-test for independent samples was used to compare differences between the two groups. Analysis of variance (ANOVA) was used to compare three groups. A linear mixed-effects regression model was used to compare the Rapid and Slow groups over time. Pearson’s χ2 test was used for categorical data analyses.

Applying the recommended effect size (Cohen’s f) 0.5 by Olink statistical services, a sample size of 44 in each of the two groups is required to detect differences at the 0.00027 significance level, with a power of 0.8.

The rate of progression to complete beta cell failure in type 1 diabetes is heterogeneous. The variation is associated with clinical characteristics, but differences in the underlying biochemistry remain largely undefined. Identifying biomarkers associated with the disease progression rate could unveil underlying mechanisms and be of prognostic value, especially when selecting suitable intervention therapies. To investigate this, we analyzed 184 biomarkers by PEA in 46 children newly diagnosed with type 1 diabetes until loss of residual C-peptide secretion and compared baseline values with 45 healthy controls. As the study was performed retrospectively, we could classify participants with type 1 diabetes as progressing rapidly or slowly.

When comparing T1D with Healthy, 81 of the 184 analyzed proteins differed significantly. This notably includes several proteins associated with type 1 diabetes pathogenesis or protection and mainly confirms earlier findings. IL-4 was lower in the T1D cohort and has been associated with both protection against autoimmunity14 as well as type 1 diabetes debut,24 25 indeed, the IL-4 receptor is expressed in pancreatic islets.26 IL-13, which is mostly known as a Th2 cytokine with anti-inflammatory properties,27 was found to be higher in the T1D cohort, which was somewhat surprising considering that previous studies have found that the IL-13 production is decreased in patients with type 1 diabetes and individuals at risk of type 1 diabetes.28 29 Also, experimental in vitro studies have found that IL-13 can reduce beta-cell apoptosis.30 CLEC4D was lower in the T1D cohort, consistent with previous research associating low CLEC4D with proinflammatory states.31 This corroborates recent findings linking significantly lower concentrations of CLEC4D with positive type 1 diabetes autoantibodies in human pancreata.13 PRDX122 is an antioxidant with immunoregulatory properties that were lower in T1D. LAP-TGF-beta 123 is the upstream pro-protein of immunoregulator TGF-beta 1 and was lower in T1D. Notably, TGF-beta 1 dissociates from LAP and becomes active on interaction with primarily integrin ITGAV:ITGB632 and can promote Th17 or Treg differentiation in a concentration-dependent manner. High concentrations of TGF-beta 1 are shown to downregulate IL-17 expression in favor of Tregs, while low concentrations can promote Th17 differentiation.33

Soluble lectin Gal-1 was lower in the T1D cohort and is an anti-inflammatory cytokine34 involved in several autoimmune diseases, including type 1 diabetes.17 In addition, studies in non-obese diabetic mice associate Gal-135 and TNFSF14 (aka LIGHT)18 with reversal of beta cell autoimmunity and insulitis, respectively.

Overall, lower levels of IL-4, Gal-1, TNFSF14, and higher levels of IL-12 in the T1D cohort are in line with a Th1-dominated immune response and previously published data. Higher IL-13 in the T1D cohort, a cytokine primarily associated with Th2 cells and immune tolerance, was unexpected and seemingly at odds with the Th1/Th2 paradigm of autoimmunity.

Our results primarily point to a lack of protective proteins in the T1D cohort compared with Healthy. In fact, cytokine levels were mostly lower in cohort T1D, possibly representing an inability to moderate an overzealous immune system attacking pancreatic beta cells. Incongruously, other biomarkers associated with type 1 diabetes pathogenesis, such as autoantibody ICA1,19 immune cell activator CD40-ligand,20 and complement activator MASP-1,21 that can be raised in type 1 diabetes, were also lower in the T1D group, illustrating the complexity and heterogeneity of the type 1 diabetes pathogenesis.

The ANOVA showed higher levels of proteins CAIX, HSD11B1, and TNFRSF21 in group Rapid versus Slow. CAIX is involved in the L-Arginine/Nitric Oxide pathway and has previously been found to be altered in children with type 1 diabetes,36 but it has not previously been associated with rapid disease progression per se. HSD11B1 is a catalyst for the reversible conversion of cortisone to cortisol and is expressed in various tissues. Nocturnal HSD11B1 has previously been found to be higher in children with type 1 diabetes37 and has been found to increase insulin resistance.38 HSD11B1 is also expressed in alpha and beta cells and is demonstrated to blunt glucose-stimulated insulin secretion.39 40 Hence, a higher HSD11B1 could speed up beta cell exhaustion via increased peripheral insulin resistance and reduced insulin secretion. This may partly explain the lower stimulated C-peptide in the Rapid group, suggesting that there could be a functional impairment of beta cells in addition to direct destruction. TNFRSF21 is involved with the negative regulation of Th2-cell activation and cytokine release.41 Upregulated TNFRSF21 and lower activation of the immune modulatory Th2-dependent pathway are in line with a more aggressive autoimmune attack in group Rapid.

In the linear mixed model, we found that none of the analyzed proteins differed in the Group term (Rapid vs Slow). However, 25 proteins, including ITGB6, IL-12, and Gal-1, differed in the Time term, suggesting an involvement in disease progression. Interestingly, ITGB6 has recently been found to be a target of autoantibodies in inflammatory bowel disease42 suggesting that it could be involved in autoimmunity.

Interestingly, in the combined Time:Group interaction term only one marker, namely CAR, was altered significantly and found to increase in the Rapid group from debut to 9 and 18 months, respectively. ITGB6 was also altered over time depending on the group, but this was not significant after adjusting for multiple testing. It is, however, striking that CAR and ITGB6 are functionally similar, being trans-membrane, ligand-binding, signaling proteins involved in cell–cell interaction and autoimmunity. CARs are present on beta cells’ surface and are a known port of entry for adenovirus and coxsackievirus, potentially triggering autoimmunity and insulitis.43 44 ITGB6 is part of a dimer receptor together with integrin alpha 5 (ITGAV) that binds coxsackievirus and is expressed in pancreatic islet cells.45 It is striking that this port of entry for viruses also activates TGF-beta 1, a known regulator of the T-helper 17 cell (Th17) pathway. Numerous epidemiological and clinical investigations support an association between enteroviruses, particularly coxsackievirus B, and autoimmune type 1 diabetes.46 47 Although the mechanism remains incompletely understood, experimental findings are suggestive of molecular mimicry or bystander T-cell activation.47 In this study, the change in CARs is specific to patients with rapidly progressing type 1 diabetes, which suggests an association between CARs and disease progression.

A strength of this study is the collection of initial blood samples shortly after type 1 diabetes diagnosis, unlike the wider definition of new-onset type 1 diabetes, which in some studies may extend up to a year after diagnosis. A limitation of the study is the inherent difficulty in interpreting how cytokine levels in peripheral blood mirror the immune response local to the pancreas. However, this approach does not compromise our aim to find prognostic biomarkers of type 1 diabetes progression rate. Furthermore, it is a limitation that 12 of 184 proteins (eg, IL-35, interferon, tumor necrosis factor) failed QC of the normalization between assays and were excluded from analyses. However, these biomarkers are deemed unlikely to separate the overlapping PCA plots. Also, the sample size of the study could limit the detection of more discrete differences between the groups.

In summary, the main aim of our study was to find biomarkers predictive of C-peptide decline in children with new-onset type 1 diabetes, but of the analyzed biomarkers, we found no clear separation in patterns that could separate the Rapid, Slow, and Healthy groups at baseline. However, we did find some markers which may be useful in predicting the decline of C-peptide. Also, these could potentially be important for our understanding of the type 1 diabetes pathogeneses, which merits further investigation.