Introduction

Type 1 diabetes (T1D) is a chronic autoimmune disease where the destruction of pancreatic beta cells leads to gradually diminishing insulin production1. Different environmental and genetic factors play an essential role in the development of the disease1. Involvement of viruses, especially enteroviruses (EV), has been linked with the development of T1D in children2, as well as with various immunological changes, including pancreatic beta cell specific reactions3. Among the latter, diabetes-associated autoantibodies (DAA), like glutamic acid decarboxylase autoantibodies (GADA), zinc transporter 8 autoantibodies (ZnT8A), islet antigen-2 autoantibodies (IA-2A) and insulin autoantibodies (IAA), occupy an essential position, since they develop months or even years before the diagnosis of T1D (stage 3)1. Most stage 3 individuals have one or more DAA with symptoms, beta cell loss and hyperglycaemia, but up to 15% of individuals with T1D are DAA-negative4. Individuals with T1D also have an increased risk for developing other autoimmune diseases. According to Hughes and co-workers, at least one other autoimmune disease was diagnosed in 27% of participants with T1D, thyroid and coeliac diseases being the most common (20% and 6%, respectively)5.

T1D susceptibility is affected by more than 50 genetic loci6, including the HLA class II region. The DR3-DQ2 and DR4-DQ8 haplotypes show the highest susceptibility7. Immunologically substantial non-HLA genes like protein tyrosine phosphatase non-receptor type 22 (PTPN22 rs2476601), cytotoxic T lymphocyte associated antigen 4 (CTLA4 rs3087243), interferon induced with helicase C domain 1 (IFIH1 rs1990760) and beta-cell functional gene solute carrier family 30 member 8 (SLC30A8 rs13266634) have been linked with T1D in several studies, but the data varies among populations8,9,10,11. Additionally, several immune checkpoint molecules have recently been proposed as prognostic markers12. Gene expression signatures of whole blood cells may predict the rate of T1D progression as one could see from the recent study of INNODIA Consortium13.

T1D is not only a childhood disease but can develop at any age1. Despite this, most previous studies have mainly focused on younger individuals with T1D. However, the demonstration of various phenotypes of childhood- and adulthood-onset T1D implies the involvement of diverse risk factors14,15. Combining phenotypic and genotypic patient characteristics has been proposed to improve the subclassification of diabetes16.

Our study set several aims. Firstly, we aimed to characterise the demographic, clinical, immunological and immunogenetic profiles associated with T1D in two age groups from Estonia, a country with relatively low T1D incidence compared to the neighbouring Nordic countries with high T1D incidence2. Secondly, we aimed to find risk factors for single and multiple DAA positivity in childhood- and adulthood-onset participants with T1D, as well as to characterise individuals with T1D who tested negative for DAA. In addition, we investigated changes in co-stimulation pathway-associated gene expression levels at diagnosis and one year after diagnosis.

Results

Baseline characteristics of participants

In the adult group the pre-diagnosis duration of symptoms was longer (P = 0.003), there were significantly more participants with weight loss (P = 0.001) and more male participants (P = 0.013) (Table 1).

Table 1 Characteristics of participants with type 1 diabetes by the age groups.
Full size table

GADA was the most prevalent autoantibody (80.1%) in the all participants, followed by ZnT8A (73.2%) and IA-2A (65.5%). Nearly half of the participants tested positive for all three DAA (45.8%), most of whom were children (Table 2). Twenty-eight (16.7%) participants were positive for one DAA and they were mostly adults, with GADA being the most prevalent single DAA in these individuals. Among the three detected DAA, the most prevalent pair combination was GADA and ZnT8A (15.5%). However, children were substantially more positive for IA-2A and ZnT8A than adults, who did not have these antibody combinations. To better monitor changes during early childhood, we compared participants in smaller age groups at 5-year intervals. As shown in Fig. 1a, participants aged < 15 years had significantly more ZnT8A and IA-2A (85.4% and 77.3%, respectively) compared with participants aged > 15 years (50.0% and 43.1%, respectively) (both P < 0.001). There were more participants positive for single DAA and less participants positive for ≥ 2 DAA in the participants aged > 15 years compared with participants aged < 15 years (both P < 0.001) (Fig. 1b). Median DAA levels (U/ml) were similar between the children and adults (Table 2). The prevalence of other detected antibodies (Table 1) was similar for children and adults. The most prevalent non-DAA autoantibody was IgG-type autoantibodies to thyroid peroxidase (anti-TPO IgG) (8.3%), followed by IgA-type autoantibodies against tissue transglutaminase (anti-tTG IgA) (4.8%). Sixty-eight participants (40.5%) were positive for IgA type antibodies against EV (anti-EV IgA) and anti-EV IgG antibodies. Two children had IgA deficiency.

Table 2 Distribution of DAA for the participants with T1D by the age groups.
Full size table
Fig. 1
figure 1

Prevalence of (a) different DAA, (b) DAA count and (c) HLA-DR haplotypes in different age subgroups in the group of participants with type 1 diabetes.

Full size image

The distribution of HLA genotypes was different between children and adults (P = 0.031) (Table 1). The most prevalent HLA genotype in children was DR3/DR4, which was the least common in adults. Children diagnosed with type 1 diabetes before the age of 5 years belonged mainly to the HLA-DR3/DR4 group. In children aged 6–10 years old, the distribution of genotypes was more even (Fig. 1c). The distribution of other detected SNP genotypes was similar between children and adults (supplementary Table S2).

Description of DAA-negative participants

The prevalence of DAA negativity was 3.6% (6 out of 168) (Table 3), with a median age of 40.6 years (IQR 34.6–42.6). Ketoacidosis occurred in 33.3% and ketonuria in 83.3% of the DAA-negative participants. Median C-peptide level was 0.12 nmol/l [inter quartile range (IQR) 0.10–0.27)], median haemoglobin A1c (HbA1c) level was 13.1% (IQR 12.8–14.4), and median BMI was 22.5 kg/m2 (IQR 20.1–24.1). Two individuals had a family history of T1D. HLA-DR3/DR4, DR4/x and DR3/x were detected in 16.7% of the DAA-negative participants. The median age of DAA-negative participants was higher [40.6 (IQR 34.6–42.6) vs. 11.9 years (IQR 7.9–20.6)], and the duration of symptoms lasted longer [82 (IQR 52–157) vs. 28 days (IQR 14–45)] compared to DAA-positive participants (P < 0.001 and P = 0.015, respectively). These six DAA-negative participants were followed by doctors and they remained DAA-negative at follow-up visits.

Table 3 Description of participants with type 1 diabetes negative for diabetes-associated autoantibodies (GADA, IA-2A, IAA and ZnT8A).
Full size table

Comparison of single DAA-positive versus multiple DAA-positive T1D

The description of the different characteristics of single DAA-positive versus multiple DAA-positive T1D is presented in Supplementary Table S3. Single DAA-positive participants’ median age was 21.2 years, whereas multiple DAA-positive participants’ median age was 11.6 years. In univariate logistic regression analysis, older age increased the odds of being single DAA-positive (OR 1.05; 95% CI 1.02–1.09, P < 0.001) versus being multiple DAA-positive. HLA DR4/x decreases the odds of being single DAA-positive (OR 0.30; 95% CI 0.09–0.98, P = 0.047). There was a tendency for more frequent anti-EV IgG positivity (45.1%) in multiple DAA-positive participants compared with single DAA-positive participants (24.0%). After adjustment for age, sex, HLA haplotypes and family history of T1D in multivariate logistic regression analysis, no significant associations were detected.

Comparison of multiple DAA-positive adults versus multiple DAA-positive children

The description of the different characteristics of multiple DAA-positive adults versus multiple DAA-positive children is presented in Supplementary Table S4. There were 30 adults with multiple DAA and 107 children with multiple DAA. Multiple DAA-positive adults were predominantly males. In multivariate logistic regression analysis adjusted for sex, HLA haplotypes and family history of T1D, anti-EV IgG positivity was a risk factor for multiple DAA positivity [adjusted odds ratio (adOR) 4.42; 95% CI 1.62–12.04, P = 0.004].

Gene expression comparison between two time points

Linear mixed models were used for analysis of gene expression levels. We did not detect a considerable difference between the two time points (1) i.e. at diagnosis and (2) one year after diagnosis on gene expression levels (Supplementary Fig. S1) in our study population. At the 1st time point, the gene expression levels of more than half (13 out of 25) of the participants revealed at least one outlying value. One year after diagnosis, the outlying values were lower and distributed more evenly. However, the forkhead box P3 (FoxP3) expression was increased among most children (Supplementary Fig. S2) at the second study point and correlated with different activation markers expression in these patients (Supplementary Fig. S2).

Some correlations with autoantibody data were also found. Using Spearman’s rank correlation, we found that at diagnosis the expression level of T cell immunoreceptor with immunoglobulin and immunoreceptor tyrosine-based inhibitory motif domains (TIGIT) was positively but weakly correlated with GADA level (P = 0.036, rho = 0.43). In contrast, levels of CD28 and FoxP3 correlated with ZnT8A level (P = 0.040, rho = 0.46 and P = 0.040, rho = 0.46, respectively). One year after diagnosis, expression level of CD155 correlated with ZnT8A level (P = 0.050, rho = 0.46). After applying the Benjamini-Hochberg’s false discovery rate FDR correction, these correlations were not significant.

Discussion

This study characterised the demographic, clinical, immunological and immunogenetic profiles associated with T1D in recently diagnosed children and adults, and described age-specific risk factors for single and multiple autoantibody positivity in a cross-sectional sample from Estonia, a country with relatively low T1D incidence, unlike the neighbouring Nordic countries. In general, the risk factors for single versus multiple autoantibody positivity between children and adults were similar. We found that EV antibodies increase the odds to have multiple DAA positivity also in adulthood-onset T1D, similar to childhood-onset T1D. Particular attention was paid to the subgroup of individuals without DAA. By measuring a broad spectrum of DAA, we showed the importance of detecting ZnT8A and IAA in adults, especially in identifying DAA-negative individuals with T1D. We found the absence of DAA to be characteristic of subtler progression of T1D. In addition, we investigated changes in co-stimulation pathway-associated gene expression levels at diagnosis and one year after diagnosis. The time point closer to diagnosis provide more unique information for evaluating the different pathway-related endotypes.

Similar to previous reports, we showed that multiple DAA autoantibodies are more characteristic of younger children with recently diagnosed T1D compared to older individuals and suggestive of a more aggressive progression of the disease. We also confirmed earlier findings17 that the duration of symptoms before diagnosis increased with age, not only because parents are more alert to the health issues of their child, but also due to the fact that the onset of the disease is less aggressive in adults. Thyroid autoimmunity was the most common concomitant autoimmune disease, as reported also by Hughes et al.5.

GADA was the most common autoantibody in our study population, which is consistent with Redondo et al.18 who showed that GADA is present in up to 80.1% of individuals with T1D. Interestingly, the most prevalent DAA combination was GADA and ZnT8A. The prevalence of ZnT8A was even higher than that of IA-2A, emphasising the importance of detecting ZnT8A in diabetes19,20. ZnT8A is related to the metabolic state of diabetes and is more often detected in younger individuals with T1D with ketoacidosis19, which can be helpful in prognosing the progression of diabetes in these participants20. The DAA profile was age-dependent, with ZnT8A and IA-2A being more prominent in participants younger than 15 years and GADA becoming more prevalent in older participants, matching the results obtained in earlier studies17,19,21. Younger age, especially under 5 years, showed increased risk for multiple DAA positivity, which supports the notion that in young children with type 1 diabetes, autoimmune beta cell destruction is more intense19,21. Additionally, as expected, participants younger than 5 years belonged mainly to the HLA-DR3/DR4 group which was less represented in older participants22,23. This result adds further evidence about a more aggressive autoimmune attack on pancreatic beta cells in younger participants21,24. This may be related to the inadequacy of the immune system’s regulation, but also to anatomical and physiological differences in the pancreas or some other undetected pathological factors in these children, as information about the immunology of the exocrine pancreatic tissue in these patients is relatively scanty.

Although EV infections have been associated with the development of T1D in young children25, we showed that EV antibodies increase the odds to have multiple DAA positivity also in adulthood-onset T1D, suggesting that repeated or prolonged EV infections are involved in T1D development in adults.

We paid particular attention to the subgroup of participants without DAA. We showed that it is crucial to examine IAA in these individuals, as one fourth of the GADA-, IA-2A- and ZnT8A-negative participants had IAA. The participants of the DAA-negative subgroup were all male adults, which is in accordance with earlier reports showing females to be less likely autoantibody negative17,26. Presence of DAA is associated with a more rapid illness progression, while the absence of DAA can point to subtler progression4. This is in line with the older age of our DAA-negative individuals and with the fact that only two of them had ketoacidosis. The subtler progression of their disease was also supported by the longer duration of symptoms before diagnosis and higher HbA1c values. Several previous studies have described the clinical profile of DAA-negative new-onset T1D in adults17,27,28. However, only a few authors measured ZnT8A, making it challenging to extrapolate their results to our study17. In our DAA-negative participants with HLA-DR3 or DR4 risk alleles, the autoimmune pathogenesis of diabetes can be more easily explained, as these risk alleles have been associated with presence of DAA4,27. Of the DAA-negative participants, 50% did not have HLA-DR3 or DR4 risk alleles. These participants could be considered as individuals with idiopathic type 1b diabetes described previously in persons of European ethnicity17,27. Another possibility is that these participants might have novel diabetes-related autoantibodies that have only recently been discovered or have not yet been demonstrated4,29,30. We can also rule out maturity-onset diabetes of the young (MODY), as the median age of our DAA-negative participants was 40.6 years, all received insulin, and their median HbA1c value was high (13.1%)31,32.

Although we demonstrated variability in the blood samples’ mRNA expression levels between the two time points, the differences were statistically not significant. Instead, we detected more outlier values in the samples taken at diagnosis compared to those taken one year later, suggesting that the time point closer to diagnosis is more informative for evaluating the different pathway-related endotypes and thereby influencing subsequent clinical treatment16,33,34,35. We agree with Suomi et al.13 who stressed the importance of individual variability in the disease course as well as peripheral blood genes expression among participants with T1D.

This study has several strengths. Firstly, the blood samples were obtained from participants up to 7 days after diagnosis. Secondly, in addition to the three T1D children age groups, divided on the basis of presumable pre-pubertal, pubertal and post-pubertal age, we studied also adults. Therefore, we were able to delineate the age-specific immunological and immunogenetic disease subtypes among the recent-onset cases. Thirdly, we assessed IAA in DAA-negative and single DAA-positive participants to state with full confidence that these individuals are who we define them to be. On the other hand, a limitation of our study was that we could not determine IAA in all participants. It would have been interesting to see what the prevalence of IAA would have been in individuals positive for multiple antibodies. Another important limitation is the relatively small sample size, which might have raised the possibility of a type II error occurring. The lack of baseline and clinical data, such as BMI, HbA1c, and C-peptide, as well as data on gene expression levels for some participants, can also be considered as a limitation of the study.

Overall, our findings suggest that with increasing age, environmental factors become more important in the development of T1D. Specifically, we observed that T1D risk haplotypes were less common in adults, suggesting a potential shift in the underlying mechanisms contributing to the disease across different age groups. Additionally, older age was associated with a higher likelihood of being single DAA positive, while exposure to EV infections was linked to an increased risk of multiple DAA positivity also in adults. These results highlight the complex interplay between genetic predisposition and environmental triggers in T1D development, particularly in older populations. In the future, besides extensive exposome studies, comprehensive research into the metagenomic and quantitative trait locus (QTL) analysis of immune cell genes is needed to characterise the age-dependent subtypes of T1D. This will most likely lead to the development of effective treatment modalities for the whole spectrum of clinical variabilities in T1D.

Methods

This cross-sectional study included 168 participants with recently diagnosed T1D (aged 2.9–68.2 years, median age 12.1 years, 56.0% men) of Europid origin. The diagnosis of T1D was based on the internationally approved diagnosis criteria36. Peripheral blood samples were obtained on the day of diagnosis or up to 7 days after that (1st time point). Follow-up blood samples (2nd time point), collected about one year after the diagnosis (318–482 days), were available for 25 children (aged 3.6–14.5 years, median age 10.3 years, 40% boys). Twenty-one out of these 25 participants at the 1st time point have been described previously37,38. The participants were recruited during 2004–2020 from the Children’s Clinic of Tartu University Hospital (Tartu, Estonia), Tallinn Children’s Hospital (Tallinn, Estonia), Internal Medicine Clinic of Tartu University Hospital (Tartu, Estonia) and Kuressaare Hospital (Kuressaare, Estonia). All participants received insulin treatment at the time of sampling. The levels of plasma C-peptide (random), glucose or HbA1c were measured by hospital’s reference laboratories according to their routine protocol. None of the recruited individuals had a history of infection during the past month, except for two participants with respiratory infections. Twenty-two participants (13.1%) had concomitant autoimmune diseases: 14 (8.3%) had thyroiditis, six (3.6%) had coeliac disease, two (0.6%) had vitiligo, and two (0.6%) had rheumatic disease.

Variables

Baseline characteristics were obtained during recruitment, including sex, age, data about the family history of T1D, and concomitant autoimmune diseases. The participants were divided into two study groups: children (≤ 18.9 years), and adults (≥ 19 years). Data about weight and height were also collected, but as 56% of the data was incomplete, BMI (kg/m2) was only calculated for DAA-negative participants. The clinical variables collected by using a questionnaire included the duration of symptoms (continuous variable). Ketonuria and ketoacidosis were diagnosed at the time of diabetes diagnosis. Onset of symptoms and recruitment time were classified according to the calendar months: spring (March–May), summer (June–August), autumn (September–November) and winter (December–February).

Detection of antibodies

From serum samples, GADA, IA-2A and ZnT8A were measured with commercial enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturer’s instructions (RSR Ltd., Cardiff, UK). The cut-off levels were ≥ 5 U/ml for GADA, ≥ 15 U/ml (tests up to May 2015) or ≥ 7.5 U/ml (tests after May 2015) for IA-2A and ≥ 15 U/ml for ZnT8A. IAA were only measured from GADA, IA-2A and ZnT8A negative and single DAA-positive samples using the luciferase immunoprecipitation system (LIPS) assay, as described in detail elsewhere39, using recombinant luciferase-tagged antigen (courtesy of Dr V. Lampasona). Results > 0.46 arbitrary units (AU) were considered positive. All tests had been under regular external quality control by the Islet Autoantibody Standardization Program40.

Anti-EV IgA and anti-EV IgG were analysed from serum samples, using in-house ELISA as described elsewhere2, with the common synthetic EV peptide KEVPALTAVETGAT (Storkbio Ltd., Tallinn, Estonia) as the antigen. The test results were expressed in enzyme immunoassay units (EIU) and calculated with the formula: [(sample absorbance–negative control reference absorbance)/(positive control reference absorbance–negative control reference absorbance)] × 100. The same negative and positive control samples were used in all analysis. According to previously published results, test results ≥ 15 EIU were taken as proof of presence of anti-EV2.

Anti-TPO IgG and anti-tTG IgA were analysed with a fluoro-enzyme immunoassay (FEIA) on an ImmunoCap100 (Phadia AB, Uppsala, Sweden). According to the manufacturer’s recommendations, the cut-off values for positivity were as follows: ≥ 100 IU/ml for anti-TPO IgG and ≥ 10 U/ml for anti-tTG IgA. The results indicating IgA deficiency (anti-tTG IgA result “Low RU”) were further verified by determination of total serum IgA.

Genotyping

PCR based lanthanide-labelled oligonucleotide hybridization and time-resolved fluorometry were used to determine HLA-DR/DQ genotyping, as described elsewhere41,42. Four comparison groups were formed based on the HLA-DRB1-DQA1-DQB1 haplotype combinations: (1) DR3/DR4, (2) DR4/x, (3) DR3/x and (4) x/x, where x represents haplotypes that were neither DR3 nor DR4. This classification is based on association between T1D and HLA haplotypes reported by Ilonen et al.7.

For the other genotyping, DNA was extracted from peripheral blood leukocytes using the salting-out method. PTPN22 rs2476601 (C__16021387_20) and IFIH1 rs1990760 (C__2780299_30) were genotyped using the TaqMan SNP genotyping assay (Applied Biosystems, Foster City, CA, USA) on the ABI 7000 instrument (Applied Biosystems, Foster City, CA, USA) according to the manufacturer’s protocol. CTLA4 CT60 rs3087243 and SLC30A8 rs13266634 were genotyped with a PCR–RFLP assay, as described elsewhere43,44. All investigated polymorphisms were in Hardy–Weinberg equilibrium (p > 0.05).

Gene expression analysis

Total RNA was extracted, isolated from peripheral blood and reverse transcribed, as described previously37. The co-stimulatory pathway molecules, the expression of which was measured, included B and T lymphocyte attenuator (BTLA), CD19, CD25, CD28, CD80, CD86, CD155, CD226, soluble CTLA4 (sCTLA4), full-length CTLA4 (flCTLA4), FoxP3, inducible T cell co-stimulator (ICOS), ICOS ligand (ICOS-L), glucocorticoid-induced tumour-necrosis-factor-receptor-related protein (GITR), GITR ligand (GITR-L), programmed cell death protein 1 (PDCD1), programmed death-ligand 1 (PD-L1), programmed death-ligand 2 (PD-L2), and TIGIT (see supplementary Table S1). The mRNA levels were assessed with Maxima Probe/ROX qPCR Master Mix (Thermo Scientific, Waltham, MA, USA). An ABI 7900 SDS instrument (Applied Biosystems, Foster City, CA, USA) was used, and the manufacturer’s protocol was followed. Expression levels were normalized against β-actin, and the relative expression of markers was quantitated with the 2-ΔΔCT method45. A success rate of 98% was obtained for the studied genes.

Statistical analysis

The R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria) was used for statistical analyses. With the sample size of 168 the post-hoc power was 0.90, with α = 0.05 and effect size 0.5. For the gene expression analysis in two time points, the sample size of 25 enabled us to detect an effect size of 0.34, with 0.90 power and α = 0.05. Continuous variables were expressed as median (IQR) and non-parametric data as n (%). The χ2, Fisher’s exact test or the proportion test was used for comparison of categorical variables, and the Mann–Whitney U test was used for comparison of continuous variables between the children’s and the adults group. Proportion test was employed to compare the prevalence of DAA for children and adults.

Univariate logistic regression analysis was performed, and ORs with 95% CI were calculated for comparison of (1) participants positive for single DAA versus those positive for multiple DAA; (2) adult versus children with positive for multiple DAA. Summer, as a low period of infections and the HLA genotype DR3/DR4, as high T1D risk haplotypes, were chosen as references. Age (only in comparison 1), sex, HLA haplotypes and family history of T1D were added to multivariate logistic regression analysis; adOR with 95% CI were calculated.

Spearman’s rank correlation was employed to assess correlation between gene expression levels and other continuous variables. A linear mixed model was used for gene expression analysis at two different time points with the lme4 package (R). The Benjamini-Hochberg’s FDR adjustment was used to correct multiple comparisons. The Hardy–Weinberg equilibrium of SNPs was verified using χ2 test. P values < 0.05 were considered statistically significant.