Longitudinal dynamics of fasting plasma uridine in the preclinical stage of type 1 diabetes

Abstract

Context

Elevated blood uridine levels have been reported in children with newly diagnosed type 1 diabetes (T1D); however, their relevance to disease onset remains unclear.

Objective

To explore whether changes in plasma uridine are associated with the development of T1D in at-risk individuals.

Methods

Pre- and post-T1D-onset oral glucose tolerance test plasma samples from 45 cases and matched controls in the Diabetes Prevention Trial-Type 1 trial were examined for uridine concentrations using LC–MS/MS. Cases with age > 21, BMI > 25, HbA1c > 5.7%, or interval from screening to T1D diagnosis < 1 year were excluded due to the potential confounding effects of aging, obesity, and insulin resistance on uridine homeostasis. Twenty-three matched case-control pairs were used to evaluate associations between uridine levels and T1D risk.

Results

Individuals who later developed T1D (cases) exhibited an increase in plasma uridine around the onset of T1D (5.03 ± 0.86 µM before and 6.10 ± 1.40 µM after T1D diagnosis, P = .001), whereas individuals who did not develop T1D (controls) showed a decrease (6.20 ± 1.32 µM at Visit 1 and 5.20 ± 1.33 µM at Visit 2, P < .001). The receiver operating characteristic curve indicates that baseline fasting uridine demonstrates discriminatory performance for T1D (area under the curve = 0.773). Mixed-effects modeling identifies a significant time × group interaction, with group-specific uridine dynamics evident prior to T1D onset in humans.

Conclusion

Our findings suggest that fasting uridine changes precede T1D onset and may serve as an early biomarker.

Type 1 diabetes (T1D) progresses through a preclinical phase characterized by β-cell–directed autoimmunity, a gradual decline in insulin secretory capacity, and metabolic alterations that precede overt hyperglycemia [1-3]. During this stage, subtle impairments in glucose handling and β-cell functional compensation may be detectable well before clinical diagnosis. Although established markers, such as islet autoantibodies and circulating C-peptide, provide important information regarding disease risk and β-cell reserve, they often reflect relatively late stages of β-cell decline and do not fully capture broader metabolic changes accompanying early disease progression [4, 5]. Identifying additional metabolic alterations that emerge during the preclinical phase may provide critical insight into our understanding of the pathophysiology of T1D development and the diagnosis of T1D in humans.

Uridine, a pyrimidine nucleoside, is required for RNA synthesis, protein glycosylation, and lipid metabolism [6]. In healthy individuals, blood uridine levels fluctuate daily with fasting and feeding [7]. Altered blood uridine levels have been associated with metabolic diseases, though the nature of this association is not fully understood [8]. Several studies indicate that uridine plays a pivotal role in energy homeostasis and glycemic control via both insulin-dependent and independent mechanisms [9-13]. Notably, elevated blood uridine levels have been reported in children newly diagnosed with T1D [14], suggesting a potential link between uridine dysregulation and T1D pathogenesis. However, the significance of uridine to T1D development and progression remains largely unexplored.

The Diabetes Prevention Trial-Type 1 (DPT-1) is a randomized controlled clinical trial that aimed to determine whether prevention or delay the onset of overt diabetes is possible in relatives of patients with T1D [15]. The DPT-1 study included age- and sex-matched case-control pairs who attended 2 in-person visits: an initial screening visit and a follow-up visit when case subjects were diagnosed with T1D. At each visit, participants completed a standardized 6-point oral glucose tolerance test (OGTT) with blood drawn at −10, 0, 30, 60, 90, and 120 minutes. This trial provides a unique opportunity to study changes in blood uridine levels around T1D onset. Analysis of OGTT blood samples from participants may reveal difference in fasting blood uridine and responses to glucose overload between subjects who developed diabetes (cases) and those who did not (controls). By conducting correlation analysis, trajectory profiling, and predictive modeling, we aim to address whether a distinct fasting uridine pattern precedes the onset of T1D in at-risk individuals.

Materials and methods

Study population, clinical data acquisition, and sample collection from banked specimens of DPT-1 study

Oral glucose tolerance test plasma samples from the NIDDK-sponsored DPT-1 study were requested, and 45 sex- and age-matched case-control pairs were identified by the NIDDK Central Repository (NIDDK-CR) based on the availability report. Each participant took 2 OGTTs: one at the screening visit (Visit 1) and the other at the postdiagnosis visit (Visit 2). Six standardized time points (−10, 0, 30, 60, 90, and 120 minutes) were used for blood sampling during OGTTs. Four participants did not have complete plasma samples available at all scheduled time points, so a total of 1076 plasma samples from 90 participants (45 pairs) were ultimately obtained from the NIDDK-CR. Levels of blood glucose, HbA1c, insulin, C-peptide, and insulin autoantibodies, as well as variables including BMI and sex of the participants, were obtained from the NIDDK-CR [15].

The inclusion and exclusion criteria for the participants are based on clinical characteristics at the onset of T1D, the composition of the study population, known factors that may affect blood uridine levels, and sample availability from NIDDK-CR. T1D is usually diagnosed in children and adolescents, although it can occur at any age [16]. The American Academy of Pediatrics (AAP) recognizes 21 years as the traditional upper boundary of pediatric care [17]. To maintain a predominantly pediatric cohort, individuals older than 21 years were excluded from the analysis. Before exclusion, the age range of all the participants was 3.4 to 38.8 years. After excluding age >21 years, the final age range was 3.4 to 19.5 years. Three matched pairs were excluded due to ages of 38.4, 34.9, and 38.8 years at screening. Obesity is a potential confounding factor for fasting uridine, so we excluded participants with a BMI > 25, a marker of overweight or obesity. If HbA1c is higher than 5.7%, this indicates prediabetes, a state where blood metabolite changes may have started. Also, when the interval from screening to clinical T1D diagnosis is too short, longitudinal metabolic changes may not be adequately captured. Based on these considerations, our criteria were:

Exclusion criteria:

1. Incomplete clinical data.

2. Age >21 years at screening.

3. BMI >25.

4. HbA1c > 5.7%.

5. Interval from screening to T1D diagnosis <1 year.

6. Compromised plasma quality (eg, hemolysis or lipemia).

7. Failure to meet LC–MS quality-control standards.

Inclusion criteria:

All remaining case-control samples that did not meet any exclusion criteria were included in the final analysis.

This study analyzed fully de-identified human plasma samples under existing Institutional Review Board approval with minimal risk, and all sample-handling procedures followed City of Hope ethical, privacy, and biosafety standards.

LC–MS quantification of plasma uridine

Plasma uridine was quantified using a liquid chromatography–tandem mass spectrometry (LC–MS/MS) method with an isotope-labeled internal standard. LC separation was performed in hydrophilic interaction chromatography (HILIC) mode on an Agilent 1290 UHPLC system with an Agilent 6490 triple quadrupole mass spectrometer. Chromatographic separation was performed using a Waters Atlantis Silica HILIC column (100 Å, 3 μm, 2.1 × 100 mm) and a Waters HILIC VanGuard guard cartridge (3 μm, 2.1 × 5 mm). The column was maintained at 45 °C with a flow rate of 0.45 mL/min. Mobile phase A was 0.1% (v/v) formic acid in water; mobile phase B was 0.1% (v/v) formic acid in acetonitrile. Gradient elution started at 98% B, rapidly transitioned to 50% B, and then re-equilibration. The retention time of uridine was approximately 1.3 minutes. During the gradient, the eluate was diverted to waste from 0 to 0.9 minutes and 1.7 to 5 minutes to minimize contamination of the mass spectrometer.

The mass spectrometer operated in multiple reactions monitoring (MRM) mode. The monitored mass transitions were m/z 243.1 → 110.1 for uridine (C₉H₁₂N₂O₆; CAS Number: 58-96-8, Sigma-Aldrich) and m/z 255.1 → 117.1 for the isotope-labeled internal standard (*C₉H₁₂*N₂O₆·H₂O; CAS Number: 202406-84-6, Cambridge Isotope Laboratories, Inc.). MRM parameters were optimized for each compound to achieve maximal detection sensitivity.

For sample preparation, 10 μL of human plasma was spiked with 10 μL of isotope-labeled uridine internal standard (2 ppm). This mixture was extracted with 150 μL of acetonitrile. Samples were mixed for 5 minutes and centrifuged at 15 000 rpm for 5 minutes. An aliquot (50 μL) of supernatant was transferred to a 96-well plate and diluted with 100 μL of acetonitrile containing 0.15% formic acid before LC–MS/MS processing.

The analytical method for plasma uridine quantification was validated in accordance with the U.S. Food and Drug Administration and the Clinical and Laboratory Standards Institute (CLSI) guidelines (CLSI document C62-A). The method showed a linear response over a concentration range of 0.25 to 4 μg/mL with a coefficient of determination (r²) greater than 0.999. Within this range, the assay demonstrated acceptable accuracy and robustness for quantifying plasma uridine.

Data processing and statistical analysis

Raw LC–MS/MS data were analyzed using Skyline (version 25.1) to obtain final plasma uridine concentrations. Clinical and demographic information came from files provided by the NIDDK-sponsored DPT-1.

All statistical analyses were performed in R (version 4.4.1). Data organization and graphing were conducted using the tidyverse packages (dplyr, tidyr, ggplot2).

Baseline characteristics were summarized separately for the case and control groups. Continuous variables are presented as mean ± SD and compared using Student's t-test if normality assumptions (Shapiro–Wilk test) were met, or the Wilcoxon rank-sum test if not. Categorical variables, including family and medical history, are presented as counts and percentages and were compared using Fisher's exact test. All tests were 2-sided. P < .05 was considered statistically significant.

Plasma glucose, C-peptide, and uridine were obtained at 6 OGTT time points (−10, 0, 30, 60, 90, and 120 minutes) for Visit 1 and Visit 2. Visit-to-visit changes were defined as Δ = Visit 2 − Visit 1. Within-pair comparisons accounting for one-to-one matching by PairID were performed using paired t-tests. Area under the curve (AUC) values were calculated using the trapezoidal method, and ΔAUC values were similarly compared. Associations between uridine and metabolic or clinical variables were evaluated separately for the case and control groups using Pearson correlation and linear regression. To capture interindividual heterogeneity in OGTT responses, time-series data were interpolated, standardized, and clustered using k-means clustering (stats package, k = 3). Differences in cluster membership between the case and control groups were assessed using Fisher's exact test.

Logistic regression was used to evaluate both baseline (Visit 1) and visit-to-visit changes (Δ = Visit 2 − Visit 1) in metabolic, clinical, and immunological variables. These analyses determined their ability to discriminate progression to T1D. Discriminatory performance was assessed using ROC analysis with AUC and 95% confidence intervals. Optimal cutoffs were determined using the Youden index. Longitudinal uridine responses during OGTT were analyzed with linear mixed-effects models for group, time, and their interaction.

Results

Sample selection and baseline characteristics of banked specimens from the DPT-1 study

After applying the predefined inclusion and exclusion criteria, 21 samples were excluded from the analysis. Of these, 5 participants had incomplete clinical data, 3 were aged over 21 years, 6 had BMI > 25 kg/m², 3 had HbA1c > 5.7% at screening, 2 case participants had less than 1 year between screening and clinical diagnosis, one sample showed evident plasma hemolysis, and 2 samples had insufficient LC–MS data quality. Ultimately, 23 matched case-control pairs were included in the final analysis (Fig. S1 [18]).

Baseline characteristics were measured for the 23 matched pairs. These included age, sex, race, and age at the first OGTT visit and BMI, blood pressure, heart rate, HbA1c, GAD64, ICA511, MIAA, and fasting plasma glucose at the first visit. Age at the second OGTT visit and relevant medical history were also included. Cases and controls showed no significant differences in these characteristics (Table 1). Family history profiles were similar in both groups (Table S1 [18]).

Table 1.

Baseline characteristics of case and control participants in the DPT-1 cohort

Feature	Control	Case	P
N	23	23
Gender = male (%)	15 (65.2%)	15 (65.2%)	1.000
Age at screening years (mean (SD))	9.41 (4.80)	9.37 (4.76)	.976
Age at 1st drawn years (mean (SD))	11.00 (4.76)	10.77 (4.77)	.869
Age at 2nd drawn years (mean (SD))	14.47 (4.45)	14.57 (4.38)	.938
BMI at 1st drawn (mean (SD))	17.80 (2.55)	18.01 (4.23)	.486
SBP at 1st drawn (mmHg, mean (SD))	101.39 (9.90)	106.65 (17.06)	.209
DBP at 1st drawn (mmHg, mean (SD))	60.65 (8.37)	65.96 (9.93)	.057
Pulse at 1st drawn (bpm, mean (SD))	77.65 (14.24)	78.04 (12.82)	.922
HbA1c at 1st drawn (%) (mean (SD))	5.27 (0.40)	5.22 (0.31)	.653
Fasting glucose at Visit 1 (mg/dL, mean (SD))	80.86 (12.11)	77.35 (11.99)	.333
GAD64 outcome (%)
Positive	19 (82.6%)	20 (87.0%)	1
Negative	4 (17.4%)	3 (13.0%)
ICA511 outcome (%)			1
Positive	13 (56.5%)	13 (56.5%)
Negative	10 (43.5%)	10 (43.5%)
MIAA outcome (%)			.752
Positive	6 (26.1%)	8 (34.8%)
Negative	17 (73.9%)	15 (65.2%)
Race (%)			.233
White	21 (91.3%)	20 (87.0%)
Black, not Hispanic	0 (0.0%)	3 (13.0%)
Hispanic	1 (4.3%)	0 (0.0%)
Unknown	1 (4.3%)	0 (0.0%)
HEENT disease (%)			1
Yes	4 (17.4%)	4 (17.4%)
No	19 (82.6%)	19 (82.6%)
Pulmonary disease (%)			1
Yes	3 (13.0%)	2 (8.7%)
No	20 (87.0%)	21 (91.3%)
Asthma (%)			1
Yes	3 (13.0%)	3 (13.0%)
No	20 (87.0%)	20 (87.0%)
Cardiovascular disease (%)			1
Yes	0 (0.0%)	1 (4.3%)
No	22 (95.7%)	22 (95.7%)
Unknown	1 (4.3%)	0 (0.0%)
Genitourinary disease (%)			.489
Yes	1 (4.3%)	0 (0.0%)
No	21 (91.3%)	23 (100.0%)
Unknown	1 (4.3%)	0 (0.0%)
Skin disease (%)			1
Yes	1 (4.3%)	2 (8.7%)
No	22 (95.7%)	21 (91.3%)
Surgery history (%)			.0706
Yes	2 (8.7%)	8 (34.8%)
No	21 (91.3%)	15 (65.2%)
Allergies (%)			1
Yes	5 (21.7%)	6 (26.1%)
No	18 (78.3%)	17 (73.9%)
Alcohol history (%)			1
Yes	1 (4.3%)	1 (4.3%)
No	22 (95.7%)	22 (95.7%)
Treatment (%)			.0453
Intervention	4 (17.4%)	11 (47.8%)
Observation	1 (4.3%)	3 (13.0%)
Oral insulin	13 (56.5%)	8 (34.8%)
Oral placebo	5 (21.7%)	1 (4.3%)

GAD64, glutamic acid decarboxylase 65 autoantibody; ICA511, insulinoma-associated antigen-2 (IA-2) autoantibody; MIAA, micro–insulin autoantibody.

Values are presented as mean (SD) for continuous variables and number (%) for categorical variables. P-values were calculated using Student's t-test or Wilcoxon rank-sum test for continuous variables, and Fisher's exact test for categorical variables.

Longitudinal changes in glycemic and metabolic parameters

In controls, fasting plasma glucose and HbA1c did not change between Visit 1 and Visit 2, but BMI was higher at Visit 2. In contrast, in cases, fasting glucose, HbA1c, and BMI were all higher at Visit 2 than at Visit 1 (Fig. 1A-1C).

Figure 1.

Changes in metabolic parameters between the first (Visit 1) and second (Visit 2) visits in case and control groups. Fasting glucose (BB Glu, before breakfast glucose, A), BMI (B), and HbA1c (C) were measured at Visit 1 and Visit 2 in both case and control groups. In addition, plasma glucose (D), C-peptide (E), and uridine (F) levels at the 0-min time point during OGTT were compared between the 2 visits. Statistical comparisons were performed using paired sample tests (paired Wilcoxon signed-rank test or paired t-test, as appropriate).

We next compared fasting metabolic parameters at the 0-min time point during OGTT between visits. In controls, fasting glucose and C-peptide were significantly higher at Visit 2 than Visit 1 (Fig. 1D and 1E); while fasting uridine was significantly lower (Fig. 1F). In contrast, in cases, fasting glucose was increased at Visit 2 (Fig. 1D), but C-peptide remained unchanged (Fig. 1E). Fasting uridine was increased (Fig. 1F). Notably, at Visit 1, uridine was higher in controls than that in cases, whereas at Visit 2, it was modestly but significantly higher in cases than that in controls (Fig. 1F).

When comparing visit-to-visit changes (Δ, defined as Visit 2 − Visit 1) between groups, cases exhibited significantly greater increases in fasting glucose and HbA1c than controls, while BMI did not differ (Fig. 2A-2C). Similarly, for the OGTT samples at the 0-min time point, cases again showed a significantly greater increase in fasting glucose compared with controls, but changes in C-peptide did not differ (Fig. 2D and 2E). Moreover, changes in fasting uridine were significantly greater in cases than that in controls and occurred in opposite directions (Fig. 2F).

Figure 2.

Comparison of visit-to-visit changes in metabolic parameters between case and control groups. Violin plots show the differences (Δ = Visit 2 − Visit 1) in fasting glucose (BB Glu, before breakfast glucose, A), BMI (B), HbA1c (C), and plasma glucose (D), C-peptide (E), and uridine (F) at the 0-min time point during OGTT, comparing case and control groups. Statistical comparisons were performed using paired sample tests (paired Wilcoxon signed-rank test or paired t-test, as appropriate).

Associations between fasting uridine and metabolic parameters

Circulating uridine is involved in lipid and glucose metabolism and is closely linked to adiposity [7, 19]. We next evaluated the relationship between fasting uridine levels at Visit 1 and BMI, as well as HbA1c in both cases and controls. We found that fasting uridine showed no correlations with BMI or HbA1c in either group (Fig. 3A and 3B), suggesting that the difference in baseline fasting uridine levels is not secondary to BMI or HbA1c.

Figure 3.

Correlations between uridine levels and metabolic parameters in case and control groups. Scatter plots show the correlations between plasma uridine levels at the 0-min time point during OGTT at Visit 1 and BMI (A) and HbA1c (B) at Visit 1, and uridine levels at the 0-min time point during OGTT at Visit 2 (C), in both case and control groups. Heatmaps display correlations of visit-to-visit changes (Δ = Visit 2 − Visit 1) in uridine (0 minutes), glucose (0 minutes), C-peptide (0 minutes), glucose (120 minutes), C-peptide (120 minutes), HbA1c, and BMI in the control group (D) and the case group (E). Correlation analyses were performed using Spearman correlation coefficients.

Building on these findings, we next assessed fasting uridine levels. In controls, fasting uridine levels at Visit 1 were positively correlated with fasting uridine levels at Visit 2. However, this positive correlation diminished in cases (Fig. 3C).

To further reveal the relevance of uridine to glycemic control, we generated correlation heatmaps showing associations among uridine (0 minutes), glucose (0 minutes), C-peptide (0 minutes), glucose (120 minutes), C-peptide (120 minutes), HbA1c, and BMI at Visit 1 in controls (Fig. S2A [18]) and cases (Fig. S2B [18]), as well as at Visit 2 in controls (Fig. S2C [18]) and cases (Fig. S2D [18]). The overall correlation patterns between fasting uridine and the other biochemical variables were broadly similar between cases and controls for Visit 1 but different for Visit 2. Specifically, in controls, fasting glucose showed a positive association with fasting uridine at Visit 1, which changed to a negative association at Visit 2. Conversely, in cases, fasting glucose was negatively associated at Visit 1 and positively associated at Visit 2. These shifts highlight distinct temporal association patterns between the groups.

Correlation analyses of visit-to-visit changes (Δ = Visit 2 − Visit 1) further revealed group-specific patterns. Specifically, there was an association between Δ uridine and Δ BMI (−0.10 in controls and 0.39 in cases), as well as between Δ uridine and Δ fasting glucose (−0.18 in controls and 0.21 in cases) (Fig. 3D and 3E).

Trajectories of glucose, C-peptide, and uridine in OGTTs

To examine if the cases exhibit different uridine responses to glucose overload compared with the controls, we measured uridine levels at 6 OGTT time points for 23 case-control pairs. Cases exhibited impaired glucose tolerance at Visit 2, with no intergroup difference at Visit 1 (Fig. 4A). Similarly, cases had a reduced C-peptide response at Visit 2, but no intergroup difference at Visit 1 (Fig. 4B). However, cases exhibited different uridine responses from controls at both Visit 1 and Visit 2 (Fig. 4C).

Figure 4.

OGTT profiles and area under the curve (AUC) comparisons in case and control groups at Visit 1 and Visit 2. Line plots show OGTT time-course profiles of plasma glucose (A), C-peptide (B), and uridine (C) measured at multiple time points during OGTT in case and control groups at Visit 1 and Visit 2. Data are presented as mean ± SEM. P-values indicate comparisons between case and control groups at Visit 1. Violin plots display differences in OGTT area under the curve (AUC) values (Δ = Visit 2 − Visit 1) for glucose (D), C-peptide (E), and uridine (F) between case and control groups. Statistical comparisons were performed using paired sample tests, and group comparisons were conducted using nonparametric tests as appropriate. AUC was calculated using the trapezoidal rule.

Expanding on the uridine response, uridine levels were higher at all time points at Visit 1 than at Visit 2 in controls. In contrast, in cases, uridine levels at almost all time points were lower at Visit 1 than at Visit 2 (Table 2).

Table 2.

Paired comparison of OGTT plasma measurements between case and control groups

Plasma analytes during OGTT	Control				Case				P (Case vs Control)
	Visit 1	Visit 2	Δ (V2–V1)	P (V2 vs V1)	Visit 1	Visit 2	Δ (V2–V1)	P (V2 vs V1)	Visit 1	Visit 2	Δ (V2–V1)
Glu (mg/dL)
pre 10 min	83.83 ± 8.28	87.43 ± 9.63	3.61 ± 7.73	.036	85.91 ± 8.07	104.30 ± 22.42	18.39 ± 23.56	.001	.432	.005	.011
0 min	82.91 ± 9.48	87.43 ± 9.07	4.45 ± 8.46	.022	86.57 ± 7.98	104.09 ± 22.06	17.52 ± 22.90	.001	.222	.005	.039
30 min	143.65 ± 24.45	146.87 ± 23.75	3.22 ± 28.39	.592	146.87 ± 34.66	181.57 ± 40.66	34.70 ± 54.96	.006	.725	<.001	.015
60 min	123.09 ± 28.73	143.70 ± 37.45	20.61 ± 48.11	.052	148.70 ± 41.19	246.26 ± 48.97	97.57 ± 60.09	<.001	.026	<.001	<.001
90 min	108.48 ± 27.28	126.35 ± 29.59	17.87 ± 42.75	.057	138.83 ± 38.82	272.57 ± 50.35	133.74 ± 61.13	<.001	.019	<.001	<.001
120 min	111.83 ± 35.69	109.74 ± 26.69	−2.09 ± 46.15	.830	121.26 ± 30.13	269.43 ± 49.04	148.17 ± 58.27	<.001	.415	<.001	<.001
AUC	15027.39 ± 2409.61	16339.35 ± 2601.08	1311.96 ± 3422.09	.080	17011.52 ± 3262.84	27656.52 ± 4965.69	10645.00 ± 5982.93	<.001	.040	<.001	<.001
C-Pep (ng/mL)
pre 10 min	0.90 ± 0.40	1.36 ± 0.46	0.46 ± 0.72	.006	0.89 ± 0.67	1.12 ± 0.66	0.23 ± 0.78	.164	.890	.099	.077
0 min	0.93 ± 0.39	1.30 ± 0.47	0.37 ± 0.68	.015	0.93 ± 0.74	1.11 ± 0.64	0.18 ± 0.82	.307	1.000	.160	.179
30 min	3.90 ± 1.84	4.19 ± 1.46	0.29 ± 2.24	.546	3.01 ± 1.28	2.17 ± 1.03	−0.84 ± 1.26	.004	.114	<.001	.025
60 min	4.24 ± 1.53	5.38 ± 1.75	1.19 ± 2.16	.017	3.57 ± 1.55	2.65 ± 1.07	−1.02 ± 1.13	<.001	.127	<.001	<.001
90 min	3.99 ± 1.60	5.28 ± 1.94	1.34 ± 2.35	.017	4.00 ± 1.75	2.99 ± 1.22	−1.11 ± 1.38	.001	.920	<.001	<.001
120 min	4.00 ± 1.61	4.40 ± 1.60	0.48 ± 2.49	.386	3.69 ± 1.71	3.41 ± 1.68	−0.36 ± 1.30	.210	.441	.056	.338
AUC	431.20 ± 169.73	531.48 ± 174.03	100.28 ± 230.60	.049	395.70 ± 159.26	301.76 ± 136.22	−93.93 ± 109.47	<.001	.502	<.001	<.001
Uridine (µM)
pre 10 min	6.37 ± 1.43	5.17 ± 1.34	−1.20 ± 1.10	<.001	5.41 ± 0.98	5.88 ± 1.04	0.47 ± 1.01	.035	.008	.034	<.001
0 min	6.20 ± 1.32	5.20 ± 1.33	−0.97 ± 1.07	<.001	5.03 ± 0.86	6.10 ± 1.40	1.07 ± 1.40	.001	.002	.012	<.001
30 min	6.50 ± 1.33	5.52 ± 1.47	−0.98 ± 0.93	<.001	5.70 ± 1.06	6.27 ± 1.23	0.55 ± 1.15	.036	.018	.033	<.001
60 min	6.25 ± 1.62	5.35 ± 1.58	−0.89 ± 0.97	<.001	5.28 ± 1.46	6.06 ± 0.93	0.76 ± 1.32	.013	.021	.065	<.001
90 min	6.10 ± 1.50	5.23 ± 1.54	−0.87 ± 1.48	.010	5.64 ± 1.21	6.13 ± 1.30	0.49 ± 1.21	.064	.151	.045	.005
120 min	6.00 ± 1.32	5.10 ± 1.53	−0.90 ± 1.21	.002	5.58 ± 1.33	6.07 ± 1.52	0.49 ± 1.52	.135	.121	.049	.004
AUC	811.51 ± 170.04	689.53 ± 185.99	−121.97 ± 108.06	<.001	711.26 ± 132.31	794.60 ± 141.67	83.34 ± 131.88	.006	.009	.025	<.001

Plasma glucose, C-peptide, and uridine concentrations measured during the oral glucose tolerance test (OGTT) at Visit 1 and Visit 2. Values are presented as mean ± SD. Δ represents the change from Visit 1 to Visit 2 (V2− V1). Between-group comparisons (case vs control) were assessed using paired t-tests matched by PairID. Within-group differences between Visit 1 and Visit 2 were also evaluated using paired t-tests.

To evaluate differences in metabolic response, we compared visit-to-visit changes in AUC (Δ = Visit 2 − Visit 1): Δ glucose AUC was higher in cases than controls (Fig. 4D), and Δ C-peptide AUC was lower in cases (Fig. 4E). Δ uridine AUC was also higher in cases, and the groups showed opposite directions of change (Fig. 4F, Table 2, −121.97 ± 108.06 for controls, 83.34 ± 131.88 for cases, P < .001).

To reveal heterogeneity in metabolite responses during OGTT, we examined the trajectories of each metabolite. Interpolated trajectories were standardized and clustered using k-means to identify groups of individuals with similar response patterns. Three distinct trajectory patterns were identified for glucose response at visit 1 (Fig. S3A and S3B [18]), and their distribution did not differ between cases and controls (Fig. S3C [18]). Similarly, 3 distinct patterns emerged in uridine response at Visit 1 (Fig. S3D and S3E [18]), with distributions again not differing (Fig. S3F [18]). In contrast to Visit 1, at Visit 2, glucose trajectory patterns differed markedly between cases and controls (Fig. S4A-S4C [18]). Uridine trajectory patterns remained similar between the groups (Fig. S4D-S4E [18]).

Potential use of uridine as a biomarker for T1D progression

We next evaluated how well each variable-fasting uridine, fasting glucose, C-peptide, BMI, HbA1c, GAD65, ICA512, MIAA, and age discriminates the risk of the progression to T1D. At baseline (Visit 1), forest plot analysis showed that fasting uridine had the strongest association with T1D risk among the variables (Fig. 5A; P = .0038). Notably, receiver operating characteristic (ROC) analysis further demonstrated good discriminatory performance for fasting uridine, with an AUC of 0.773 (95% CI, 0.626-0.919) (Fig. 5B). The optimal cutoff of uridine concentration was 5.739 µM (sensitivity = 0.913, specificity = 0.636). Finally, boxplots show the baseline distributions of predictors in case and control participants (Fig. 5C).

Figure 5.

Baseline (Visit 1) predictors of case-control status. (A) Odds ratios from logistic regression models evaluating baseline predictors measured at Visit 1. Error bars indicate 95% confidence intervals. (B) Receiver operating characteristic (ROC) curves showing the discriminatory performance of fasting uridine alone and a combined baseline predictor model, with AUC values indicated. (C) Distribution of baseline predictors at Visit 1 in control and case groups shown as box-and-jitter plots.

Building on the baseline analysis, we next used the same analytical framework to evaluated the discriminatory ability of visit-to-visit changes (Δ = Visit 2 − Visit 1) in selected metabolic and clinical variables–namely, fasting uridine, fasting glucose, C-peptide, BMI, and HbA1c–to predict progression to T1D. Forest plot analysis compared the magnitude of change in each variable between cases and controls, showing that changes in fasting uridine (P = .01879) and HbA1c (P = .03876) were significantly associated with T1D progression (Fig. S5A [18]). ROC analysis assessed the discriminatory performance of changes in each variable, with changes in fasting uridine yielding an AUC of 0.879 (95% CI, 0.777-0.982) (Fig. S5B [18]). Integrating changes in fasting uridine with the other variables in a combined model further improved discriminatory performance, resulting in a higher AUC of 0.966 (95% CI, 0.918-1.000) (Fig. S5B [18]). Boxplots display the distributions of visit-to-visit changes in each predictor between cases and controls (Fig. S5C [18]).

To further explore temporal patterns, mixed-effects modeling was conduct, which revealed a significant time × group interaction (Table 3). Specifically, at the screening visit, individuals who later developed T1D exhibited significantly altered uridine dynamics during OGTT compared with controls; however, these group-specific differences were no longer evident at the diagnosis visit.

Table 3.

Mixed model results for OGTT uridine responses at V1 and V2

Term	V1_Estimate	V1_SE	V1_P	V2_Estimate	V2_SE	V2_P
Intercept	6.36268	0.24522	.00000	5.29372	0.26554	.00000
Time	−0.00259	0.00129	.04517	−0.00064	0.00111	.56359
Group (case vs control)	−1.04014	0.34663	.00418	0.74891	0.37560	.05188
Time × group interaction	0.00515	0.00182	.00495	0.00135	0.00157	.39250
SD_(Intercept)	1.09985			1.22224
SD_Observation	0.70250			0.60878

Linear mixed-effects model results evaluating uridine concentrations during OGTT as a function of time, group, and their interaction at Visit 1 and Visit 2.

Discussion

Fasting uridine levels are found to be elevated in children with newly diagnosed T1D [14]. To examine whether blood uridine levels change around the onset of T1D, we conducted the current study using blood samples from the DPT-1 clinical trial. A key finding of this study is that fasting uridine levels in cases is increased around the onset of T1D, whereas those in controls is decreased. Since fasting uridine levels are lower in cases than that in controls before the onset of T1D, the increase in fasting uridine around the onset of T1D appears to account for the elevation observed in children with newly diagnosed T1D [14]. Consistently, the ROC curve indicates that fasting uridine at the screening visit has a predictive value for T1D.

The postprandial fall in blood uridine is a major factor contributing to the lower fasting uridine levels [7]. Therefore, abnormal postprandial blood uridine dynamics may lead to increased fasting uridine levels. Metabolically healthy humans lower fasting uridine levels after meals, regardless of BMI [7, 20]. In contrast, insulin-resistant individuals with morbid obesity cannot lower fasting uridine after food intake [21]. This disruption in uridine homeostasis is not corrected after bypass surgery [21], a procedure that improves systemic insulin sensitivity. A recent study reported significantly higher fasting uridine levels in patients with T2D than that in healthy individuals and suggested that fasting uridine levels are a predictor of T2D [22]. Thus, abnormal elevations in fasting uridine may be common in insulin resistance and insulin insufficiency. Notably, excess uridine may attenuate insulin sensitivity by promoting O-linked N-acetylation of glucosamine (O-GlcNAcylation) on serine/threonine residues, a posttranslational modification that interferes with insulin signaling propagation [10-12, 23]. It remains unclear whether the elevation in plasma uridine dampens the cellular insulin response in at-risk individuals, thereby aggravates T1D onset.

Preclinical T1D is associated with reduced β-cell function that leads to insulin insufficiency. In this study, a divergence between glucose and C-peptide responses was observed at Visit 2: glucose levels were increased without a corresponding rise in C-peptide, indicating impaired β-cell function in cases. It remains unknown whether elevated plasma uridine contributes to the loss of β-cell function in at-risk individuals. Inflammation likely precedes the development of autoimmunity [24]. Thus, changes in local proinflammatory cytokines are a potential trigger for β-cell functional loss. Interestingly, uridine has both pro- and anti-inflammatory effects. Uridine can activate toll-like receptor (TLR) 8 on macrophages and dendritic cells in the presence of single-stranded RNA co-ligand, leading to the release of multiple proinflammatory cytokines [25]. However, uridine has also been reported to have an anti-inflammatory effect in a rat model of lung inflammation [26]. Thus, future studies are warranted to examine the impact of uridine on local islet inflammation in preclinical models of T1D.

This study has several limitations. The sample size was modest due to availability from NIDDK-CR. Larger prospective cohort studies are needed to further establish the relationship between fasting uridine and T1D onset. This study included some clinical results provided by the NIDDK-CR [15, 27-29], but the integrational analysis for uridine and immunological markers is incomplete. Although autoantibody data for GAD65, ICA512, and MIAA are available, insulin autoantibody (IAA)—a key marker in early T1D—is not available in this cohort. Subjects with confirmed IAA ≥ 80 nU/mL showed more characteristics consistent with a higher risk of diabetes [27]. OGTT is widely used as a standard for assessing glucose tolerance in humans. However, the uridine response to glucose overload remains largely unexplored. It is unclear whether the mixed meal tolerance test reveals a uridine trajectory similar to that of the OGTT in humans. Lastly, T1D is an autoimmune disease that develops over the years. Mechanisms that initiate failure of immune tolerance to β-cells remain obscure. Mechanistic studies are necessary to clarify the biological pathways linking uridine metabolism to T1D.

Funding

This study was supported by the National Institutes of Health (R01 DK126975 and R01 DK140109, Y.D.; R01 HL137723, R01 HL156951, and R01 HL171309, Z.V.W.), Helmsley Charitable Trust (to Y.D.), American Heart Association (24TPA12988803, Y.D.), and the Lions Clubs International Foundation (an unrestricted grant to Y.D.). L.F. is supported by a Larry L. Hillblom Foundation fellowship grant.

Author contributions

Q.X. conceived of and conducted experiments with help from C.J., K.W., L.F., L.D., Z.V.W., and P.W. Y.D. conceived of and designed the study. Q.X. and Y.D. wrote the manuscript. All authors commented on and approved the manuscript.

Disclosures

There is nothing to disclose for all the authors.

Data availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.