Circulating metabolites in progression to islet autoimmunity and type 1 diabetes

Aims/hypothesis Metabolic dysregulation may precede the onset of type 1 diabetes. However, these metabolic disturbances and their specific role in disease initiation remain poorly understood. In this study, we examined whether children who progress to type 1 diabetes have a circulatory polar metabolite profile distinct from that of children who later progress to islet autoimmunity but not type 1 diabetes and a matched control group. Methods We analysed polar metabolites from 415 longitudinal plasma samples in a prospective cohort of children in three study groups: those who progressed to type 1 diabetes; those who seroconverted to one islet autoantibody but not to type 1 diabetes; and an antibody-negative control group. Metabolites were measured using two-dimensional GC high-speed time of flight MS. Results In early infancy, progression to type 1 diabetes was associated with downregulated amino acids, sugar derivatives and fatty acids, including catabolites of microbial origin, compared with the control group. Methionine remained persistently upregulated in those progressing to type 1 diabetes compared with the control group and those who seroconverted to one islet autoantibody. The appearance of islet autoantibodies was associated with decreased glutamic and aspartic acids. Conclusions/interpretation Our findings suggest that children who progress to type 1 diabetes have a unique metabolic profile, which is, however, altered with the appearance of islet autoantibodies. Our findings may assist with early prediction of the disease. Electronic supplementary material The online version of this article (10.1007/s00125-019-04980-0) contains peer-reviewed but unedited supplementary material, which is available to authorised users.


47
Type 1 diabetes (T1D) is an autoimmune disease, which arises due to the destruction of the insulin 48 producing pancreatic β-cells by the immune system 1 . The incidence of T1D is highest in children 49 and adolescents in the developed countries 2 and an increase in disease rate is expected in young 50 children aged less than 5 years 3 . To reverse the increasing rate, early prediction and prevention of 51 T1D is essential. However, the aetiology of T1D disease is complex, multifactorial, and the 52 primary cause for initiation and disease progression is poorly understood 1 . Therefore, predictive 53 and preventive measures for T1D remain unmet medical needs. 54 Human leukocyte antigen (HLA) complex alleles constitute the most relevant and the strongest 55 genetic risk factor for T1D susceptibility 4 . However, only 3-10% of the individuals with risk HLA 56 loci develop T1D 5 , indicating that exogenous factors such as environmental exposure, diet and gut 57 microbiota likely play a vital role in disease progression 6 . Initiation of β-cell autoimmunity is the 58 first detectable sign of progression towards T1D. However, seroconversion to islet autoantibody 59 positivity may not lead to overt diabetes 7 and the period between the seroconversion and the 60 appearance of clinical symptoms of T1D may vary between individuals from a few months to many 61 years 8, 9 . 62 Previous studies suggest that children who progress to T1D have dysregulated metabolic profiles 63 already in infancy 10,11,12,13 , prior to the seroconversion for islet autoantibodies. However, the 64 studies in humans have so far mainly focused on lipids, and there is relatively little information 65 on polar metabolites, such as those involved in central metabolic pathways, in relation to T1D 66 pathogenesis. Herein we study circulating polar metabolite profiles in progression to T1D in a 67 longitudinal study setting. 68

Impact of age on circulating metabolome 70
We performed metabolomics analysis of polar metabolites in plasma from 120 children, divided 71 into three study groups: those who progressed to T1D (PT1D, n = 40), who seroconverted to at 72 least one autoantibody (Ab) positivity but without clinical symptoms of T1D (P1Ab, n = 40), and 73 matched Ab negative controls (CTR,n = 40). For each participant, plasma samples were collected 74 corresponding to the ages of 3, 6, 12, 18, 24, and 36 months (Fig. 1). We detected metabolites from 75 across a wide range of chemical classes including amino acids, carboxylic acids (mainly free fatty 76 acids and other organic acids), hydroxyacids, phenolic compounds, alcohols, and sugar 77

derivatives. 78
Principal components analysis (PCA) 14 of the complete dataset including identified metabolites 79 displayed an age-dependent pattern (Supplementary information (SI) Fig.1). To resolve the 80 impact of age on plasma metabolome, we performed analysis of variance (ANOVA)-simultaneous 81 component analysis (ASCA) 15 by incorporating three factors: age, gender, study cases (CTR, P1Ab, 82 PT1D) and their interactions. As expected, age related variation displayed maximum effect (4.2 %, 83 p= 0.001) in the circulating metabolome as compared to the impact of the other two factors, 84 'study groups' (1.2 %, p = 0.001) and 'gender' (0.5 %, p = 0.002). Notably, the interaction factor 'age 85 and cases' also showed a significant effect (2.9 %, p = 0.033), while interactions between other 86 factors (age/gender and case/gender) remained insignificant (p = 0.508 and p = 0.221, 87 respectively). 88 The scores from the first principle component (PC1) of the factor 'age' clearly showed an age-89 related trajectory in the circulatory metabolites (Fig. 2). The loading revealed high levels of 90 branched chain amino acids (BCAA) in the 18, 24 and 36 month age-cohorts, whereas tryptophan, 91 3-indole acetic acid (tryptophan derivative) and carboxylic acids (mainly free fatty acids) were 92 elevated during early infancy (3 and 6 months). However, we did not detect any age-dependent 93 patterns in phenolic compounds, alcohols, hydroxyacids, and sugar derivatives (SI Fig. 2). 94

Metabolite profiles during progression to islet autoimmunity and T1D 95
Considering the age as a major confounder in the plasma metabolome, we performed age-96 matched comparisons between the three study groups (CTR, P1Ab, and PT1D). Univariate analysis 97 revealed that all major metabolite classes, including amino acids, free fatty acids and sugar 98 derivatives were altered, already in infancy, among the children who later progressed to T1D (Fig.  99 3). Altogether 15 metabolites were different between PT1D and CTR groups at three months of age 100 (nominal p-value < 0.05). Nine out of 15 metabolites were significantly lower in T1D progressors as 101 compared to controls (FDR threshold of 0.1) (Fig. 3, SI Table 1). In order to assess if gender had 102 an impact on plasma metabolite levels in children at three months of age, we carried out ASCA 103 analysis with factor: study cases and gender, and their interaction. When evaluating the statistics 104 from these factors, we found only study cases had significant effect (p = 0.012), while gender and 105 their interaction remained insignificant (p = 0.081 and p = 0.73, respectively). The score of the 106 factor 'study cases' showed distinct metabolic clusters between PT1D, P1Ab and CTR, suggesting 107 that specific metabolic changes precede islet autoimmunity and T1D. The loadings disclosed that 108 methionine, 2-ketoisocaproic acid, bisphenol A, pyruvic acid, glycerol-2-phosphate, and 109 levoglucosan were higher in the PT1D group when compared with the P1Ab and CTR groups (SI 110 Fig. 3). 111 112 At 6 months of age, altogether 20 metabolites differed between PT1D and CTR (nominal p-value < 113 0.05). Fifteen of these circulating metabolites passed the FDR threshold of 0.1 (Fig. 3a-c, SI Table  114 2), including several amino acids, sugar derivatives, free fatty acids and various other organic 115 acids. The levels of most of these metabolites decreased in T1D progressors during the same 116 period as compared to CTR. Only methionine was found increased in PT1D as compared to CTR 117 at the age of 6 months. In addition, multivariate ASCA analysis revealed that only study group 118 (CTR, P1Ab, and PT1D) had a significant effect (p = 0.004) in the plasma metabolites of 6-month-119 old children, whereas the impact of gender (p =0.180) and its interaction with study group (p = 120 0.269) remained insignificant. 121 Next, we sought to examine weather children across the three study groups had altered plasma 122 metabolite levels in the age cohorts of 12, 18, 24, and 36 months. With the exceptions of 1-123 dodecanol and glycolic acid, no other statistically significant differences between the study groups 124 were observed (FDR threshold of 0.1). At 36 months of age, dodecanol level was higher in PT1D as 125 compared to CTR. Meanwhile, glycolic acid was lower in PT1D as in P1Ab at 18 months of age. 126 However, in longitudinal series these metabolites showed inconsistent trends (Fig. 3b). 127 We also studied whether group of metabolites at early age associated with a specific metabolic 128 pathway. The altered metabolites (p < 0.05) between CTR and PT1D at 3 and 6 months of age were 129 subjected to metabolic pathway analysis (MetPA) in MetaboAnalyst 16 . In line with findings at the 130 individual metabolite levels, we found that four metabolic pathways, linoleic acid metabolism, 131 arachidonic acid metabolism, alanine, aspartate and glutamate metabolism and D-glutamine and 132 D-glutamate metabolism remained altered between PT1D and CTR groups at the age of three 133 months (Fig. 4a, SI Table 3). Similarly, at 6 months of age, MetPA revealed that alanine, 134 aspartate and glutamate metabolism, D-glutamine and D-glutamate metabolism, tryptophan 135 metabolism, arginine and proline metabolism, as well as aminoacyl-tRNA biosynthesis remained 136 dysregulated between the controls and T1D progressors (Fig. 4b, SI Table 4). 137

Metabolome before and after the first appearance of islet autoantibodies 138
In order to study the effect of islet seroconversion on metabolome, we compared metabolite levels 139 before and after the appearance of first islet autoantibody in P1Ab and PT1D groups. Pairwise 140 comparison revealed that eleven metabolites were altered by seroconversion in P1Ab (nominal p-141 value < 0.05, SI Table 5), with four passing the FDR threshold of 0.1 (glutamic, aspartic, malic, 142 and 3, 4-dihydroxybutanoic acids) (Fig. 4). We detected seven metabolites altered before and 143 after islet autoantibody appearance in PT1D (nominal p-value < 0.05), but none of these passed 144 the FDR threshold of 0.1 (SI Table 6). Metabolic pathway analysis corroborated these findings 145 and revealed that alanine, aspartate and glutamate metabolism were altered when comparing the 146 pathways before and after seroconversion within P1Ab and PT1D groups (Fig.4). However, the 147 level of impact for these pathways varied between P1Ab and PT1D, with impact values 0.441 and 148 0.176, respectively. Other relevant pathways and their impact are summarized in SI Table 7 and 149 Table 8. When examining metabolite level changes in relation to the appearance of specific islet 150 autoantibodies (islet cell antibodies (ICA), insulin autoantibodies (IAA), islet antigen 151 2 autoantibodies (IA-2A), and GAD autoantibodies (GADA)), no specific associations were 152 identified, which may be due to the small number of cases per individual autoantibody. 153

154
Our study identified specific metabolic disturbances in children who progressed to T1D, as 155 compared to their age matched controls including children who developed a single islet 156 autoantibody but did not progress to T1D during the follow-up. We found that such metabolic 157 dysregulation exists before the first signs of islet autoimmunity. In agreement with earlier 158 studies 10, 17, 18 , a strong association of the metabolome was observed with age. We identified a 159 distinct plasma amino acid profile in PT1D children, particularly at the ages of 3 and 6 months. 160 Glutamic and aspartic acids as well as tryptophan remained downregulated during the early 161 infancy in PT1D as compared to CTR, but not to P1Ab. In our previous study of polar metabolites 162 in T1D progression, we found no significant difference in different age cohorts when comparing 163 PT1D and CTR groups 13 , which may however be attributable to the small number of individuals in 164 the metabolomics part of that study. Notably, in agreement with the previous study, we also 165 observed that the appearance of islet cell autoantibodies was associated with down-regulation of 166 aspartic and glutamic acids 13 , also corroborated by observed change in alanine, aspartate and 167 glutamate metabolism in the MetPA. Our findings are consistent with previous study suggesting 168 that amino acid dysregulation precedes the appearance of islet autoantibodies and progression to 169 T1D 12 . Several free fatty acids were also downregulated at 3 months of age. During basal metabolic 170 processes, triglycerides (TGs) are broken down to fatty acid and glycerol 19 . Fatty acid act as an 171 important fuel source for cells, which is required to maintain systematic energy homeostasis 20 . 172 Usually, under conditions when the availability of carbohydrate is not enough, the fatty acids are 173 alternative substrate for energy production 21 . Here, fatty acid decrease may be an indication of 174 increased energy demand in PT1D, further substantiated by the diminishment of circulating sugar 175 derivatives as well as altered linoleic acid metabolism and arachidonic acid metabolism. This is 176 also in line with our previous report 10 associating downregulated TGs and phospholipids in the 177 PT1D group, supporting the view that altered energy metabolism is involved in the initiation of 178 the autoimmune process and T1D. 179 Accumulating evidence suggests that perturbations in the gut microbial structure are associated 180 with, and contribute to the pathogenesis of β-cell autoimmunity and to overt T1D 22, 23, 24 . Here we 181 found that 4-hydroxyphenyllactic acid 25, 26 , 11-eicosenoic acid 27 , and succinic acid 28 , the 182 metabolites of potential microbial origin (catabolites), are significantly downregulated at early 183 age (3 and 6 months) in PT1D. The tryptophan derived microbial catabolite 3-indoleacetic tended 184 to be also downregulated in PT1D (SI Fig. 4). Catabolites generated by the gut microbes are vital 185 to the intestinal homeostasis 26, 29 , thus it is likely that scarcity of substrates for microbial 186 catabolism contribute to the dysbiosis associated with progression to T1D. 187 While most of the amino acids were downregulated in PT1D as compared to CTR and P1Ab, 188 methionine remained persistently upregulated in T1D progressors. This appears to be in 189 disagreement with previous studies in BABYDIAB and MIDIA cohorts, which showed decreased 190 level of methionine in autoantibody positive individuals and T1D progressors, respectively 18, 30 . 191 This discrepancy may however be explained: (1) BABYDIAB study compared children 192 seroconverting early in life (≤2 years) to those who developed autoantibodies at older age, while 193 (2) MIDIA study highlighted differences, which were mainly linked to the age of the children and 194 the duration of breastfeeding 30 . We performed similar comparison to that of BABYDIAB in the 195 current study setting but found no significant differences between the groups compared. The The ASCA multivariate analysis revealed that plasma BPA was upregulated in PT1D group, 202 although univariate analysis across different age cohorts did not reveal significant changes 203 between the groups. Studies in experimental model of autoimmune diabetes suggest that 204 increased BPA exposure is associated with accelerated development of autoimmune diabetes 33,34 . 205 However, we consider that at the present stage our findings on the association of BPA and T1D are 206 still inconclusive, because (1) in our study setting we could not control for the effect of sample 207 storage on the plasma BPA levels and (2)   Taken together, while confirming several earlier findings, the present study highlights the 211 importance of core metabolic pathways such as amino and fatty acid metabolism in early 212 pathogenesis of T1D. Metabolites of microbial origin were also found associated with T1D 213 progression. We also observed that appearance of islet autoantibodies does have an effect on the 214 amino acid levels, specifically on glutamic and aspartic acids. However, these changes do not 215 seem to be specifically associated with T1D but are instead a general feature of islet autoimmunity 216 -suggesting that amino acid imbalance may be a contributing factor in the initiation of 217 autoimmunity 13 . Our study also indicates that the largest metabolic changes associated with T1D 218 progression occur already in early infancy, then these early metabolic signatures become less 219 pronounced or even disappear with age, particularly after the initiation of islet autoimmunity. 220

This may have important implications in the search of early metabolic markers of T1D and for 221
understanding the disease pathogenesis. 222

223
These methods are expanded versions of descriptions in our related work 10 . 224

Study setting 225
The plasma samples were from the Finnish Type 1 Diabetes Prevention and Prediction Study 226 (DIPP) 35 . The DIPP study has screened more than 220,000 newborn infants for HLA-conferred 227 susceptibility to T1D in three university hospitals (Turku, Tampere, and Oulu) in Finland 36 . The 228 subjects in the current study were from the subset of DIPP children from the Tampere study 229 centre. The ethics and research committee of the participating university hospital approved the 230 study protocol and the study fallowed the guidelines of the Declaration of Helsinki. Parent for all 231 participants signed written informed consent at the beginning of the study. We collected five 232 longitudinal samples per child, corresponding to either of the ages of 3, 6, 12, 18, 24, and 36. This 233 longitudinal cohort comprises of samples from 120 children: 40 progressors to T1D (PT1D), 40 who 234 tested positive for at least one Ab in a minimum of two consecutive samples but did not progress 235 to clinical T1D during the follow-up (P1Ab), and 40 controls (CTR) subjects who remained islet 236 autoantibody negative during the follow-up until the age of 15. We matched the participants in 237 the three study group for HLA-associated diabetes risk, gender and period of birth. In total, we 238 collected 415 non-fasting, blood samples. We separated plasma within 30 minutes after the blood 239 collection by centrifugation at 1600g for 20 minutes at room temperature. The plasma samples 240 were stored at -80°C until analysed. 241

HLA genotyping 242
HLA-conferred susceptibility to T1D was analysed using cord blood samples as described by Alanine, Stearic acid, Linoleic acid, Palmitic acid and Oleic acid. Standards were purchased from 281 Sigma-Aldrich (Steinheim, Germany) and dissolved in methanol. Calibration curves included at 282 least six concentration points in the range from 1 ng/sample up to 3000 ng/sample, depending on 283 the abundance in plasma. R 2 was from 97.1% up to 99.9%. 284 Derivatization was performed instrumentally using MPS2 (Gerstel; Mülheim an der Ruhr, 285 Germany) with two robotic hands guided by Maestro software. Samples were evaporated to 286 dryness before two-step extractions. In the first step 25 μl of methoxyamine hydrochloride (TS-287 45950; Thermo Scientific: USA) was added to the sample. While mixing, the solution was 288 incubated for one hour at 45 °C. In the second step, 25μl of N-methyl-N-289 trimethylsilyltrifluoroacetamide (Sigma-Aldrich; Steinheim, Germany) was added. Incubation was 290 again performed for one hour at 45 °C. Before injection 50 μl of hexane was added to increase the 291 volatility of the solvent. Additional standards here added during derivatization. n-alkanes (c = 8 292 mg/l in MSTFA) were used for calculation of retention indexes and 4,4′-293 dibromooctafluorobiphenyl (c = 9.8 mg/l in hexane) were used as syringe standard to control the 294 quality of injection. 1 μl of derivatized sample was injected after derivatization program was 295 completed. 296 Derivatised compounds were analysed using Pegasus 4D system (LECO; Saint Joseph; USA). 297 Method used is based on two-dimensions gas chromatography followed by high speed time of 298 flight acquisition of EI fragmented mass spectra. Primary column was 10 m × 0.18 mm I.D. Rxi-5 299 ms (Restek Corp., Bellefonte, PA, USA) and secondary column 1.5 m × 0.1 mm I.D. BPX-50 (SGE 300 Analytical Science, Austin, TX, USA). System was guarded by retention gap column from 301 deactivated silica (1.7m, 0.53 mm ID, FS deactivated, Agilent technologies, USA). Modulator used 302 nitrogen gas which was cryogenically cooled. Second dimension cycle was 4s. Temperature 303 program started with 50 °C (2 min) then a gradient of 7°C up to 240°C was applied and finally 304 25°/min to 300 °C where it was held stable for 3 min. Temperature program of secondary column 305 was maintained 20 °C higher than the primary column. Acquisition rate was kept on 100 Hz. 306 Instrument was guided by ChromaTOF software (version 4.32;LECO Corporation,St. Joseph,307 USA) which was also used calculating area under the peaks with SN>100 and potential 308 identification of peaks using NIST14 and in-house library. Processing method included calculation 309 of retention indexes. Selected compounds were quantified against external calibration curves. 310 Results were exported as text files for further processing with Guineu 42 software. 311

Data analysis 312
All statistical analyses were performed on log-transformed intensity data. The transformed data 313 were mean cantered and auto scaled prior to multivariate analysis. The multivariate analysis was Wilcoxon rank-sum test was performed for comparing the two study groups of samples (e.g. PT1D 320 vs. P1Ab) in a specific age cohort. For comparison, one sample per subject, closest to the age 321 within the time window, has been used in each test. Paired t-test was performed for the matched 322 groups of samples (e.g. before vs. after seroconversion). The resulting nominal p-values were 323 corrected for multiple comparisons using Benjamin and Hochberg approach 43 . The adjusted p-324 values < 0.1 (q-values) were considered significantly different among the group of hypotheses 325 tested in a specific age cohort. All of the univariate statistical analyses were computed in 326 MATLAB 2017b using the statistical toolbox. The fold difference was calculated by dividing the 327 mean concentration of a lipid species in one group by another, for instance mean concentration during compound name matching were excluded from the subsequently pathway analysis. We 335 implemented Globaltest hypergeometric testing method for the functional enrichment analysis. 336 The pathway topological analysis was based on the relative betweenness measures of a metabolite 337 in a given metabolic network and for calculating the pathway impact score. Based on the impact 338 values from the pathway topology analysis the impact value threshold was set to > 0.10. 339

Data availability 341
The metabolomics data and the associated meta-data are deposited at the MetaboLights database 342 46 with the acquisition number (MTBLS802). All the data supporting the findings of this study 343 are available from MetaboLights database or from the corresponding authors on reasonable 344 request. 345

346
This work was supported by the JDRF grants 4-1998-274, 4-1999-731 4-2001-435   who progressed to T1D (PT1D), who seroconverted to one islet autoantibody but did not progress 521 to T1D during the follow-up (P1Ab), and control (CTR) subjects who remained islet autoantibody 522 negative during the follow-up until the age of 15 years. For each child, longitudinal plasma 523 samples were drawn, corresponding to the ages of 3, 6, 12, 18, 24, and 36 months. In each age 524 cohort and study group, number of autoantibody positive children is marked and represented 525 with Y-shape. 526 Figure 2. PCA score plots of the factor age, based on ASCA. These scores represent the 527 metabolomics dataset arranged according to the age in the PCA score plot. Each sample is 528 represented by a point and coloured according to the age. The ages of the participants are marked 529 on the x-axis while y-axis represents the sample score. Samples with similar score cluster 530 together. 531 Pathway impact values were calculated from pathway topology analysis using MetaboAnalyst. 543