Serum Metabolomics Reveals a Potential Benefit of Methionine in Type 1 Diabetes Patients with Poor Glycemic Control and High Glycemic Variability

Glycemic variability (GV) in some patients with type 1 diabetes (T1D) remains heterogeneous despite comparable clinical indicators, and whether other factors are involved is yet unknown. Metabolites in the serum indicate a broad effect of GV on cellular metabolism and therefore are more likely to indicate metabolic dysregulation associated with T1D. To compare the metabolomic profiles between high GV (GV-H, coefficient of variation (CV) of glucose ≥ 36%) and low GV (GV-L, CV < 36%) groups and to identify potential GV biomarkers, metabolomics profiling was carried out on serum samples from 17 patients with high GV, 16 matched (for age, sex, body mass index (BMI), diabetes duration, insulin dose, glycated hemoglobin (HbA1c), fasting, and 2 h postprandial C-peptide) patients with low GV (exploratory set), and another 21 (GV-H/GV-L: 11/10) matched patients (validation set). Subsequently, 25 metabolites were significantly enriched in seven Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways between the GV-H and GV-L groups in the exploratory set. Only the differences in spermidine, L-methionine, and trehalose remained significant after validation. The area under the curve of these three metabolites combined in distinguishing GV-H from GV-L was 0.952 and 0.918 in the exploratory and validation sets, respectively. L-methionine was significantly inversely related to HbA1c and glucose CV, while spermidine was significantly positively associated with glucose CV. Differences in trehalose were not as reliable as those in spermidine and L-methionine because of the relatively low amounts of trehalose and the inconsistent fold change sizes in the exploratory and validation sets. Our findings suggest that metabolomic disturbances may impact the GV of T1D. Additional in vitro and in vivo mechanistic studies are required to elucidate the relationship between spermidine and L-methionine levels and GV in T1D patients with different geographical and nutritional backgrounds.


Introduction
Type 1 diabetes (T1D) is characterized by intrinsic immune destruction of β-cell, and insulin-dependent treatment to achieve euglycemia results in substantial glycemic variability (GV) [1]. In patients with T1D, an accurate and complete GV assessment is critical for evaluating recent glucose control and the long-term risk of chronic complications. Although glycated hemoglobin (HbA1c) has been recognized as the "gold standard" for assessing glycemic control, it does not reflect intra-and interday GV that gives rise to acute events or postprandial hyperglycemia related to both microvascular and macrovascular complications. Improving HbA1c significantly reduces the risk of target-organ complications [2,3]. However, macrovascular complications continue, albeit at lower rates [4], which might be related to GV. Given the close correlation between GV and diabetic complications and hypoglycemia, the international consensus on the use of continuous glucose monitoring measured using the electronic stadiometer and weighing scale. BMI was calculated as weight in kg/(height in m) 2 . Patients were asked to stand with their arms crossed on the contralateral shoulders for waist circumference measurement. The measuring tape should be snug around the lateral aspect of each ilium at the mid-axillary line. The patients were asked not to smoke or drink strong tea or coffee within 30 min before blood pressure measurement, and to empty their bladder. Blood pressure was measured using cuff pressurization after sitting in a chair with a backrest in a quiet environment for at least 5 min. Fasting venous blood samples were drawn from all study participants in the morning after they had fasted for 10 h for the following tests: HbA1c, FBG, FCP, and lipid profiles. Blood samples for 2hBG and 2hCP testing were taken 2 h following a mixed-meal tolerance test (MMTT; 44.4% carbohydrates, 47.7% fat, and 7.9% protein).
Automated high-performance liquid chromatography was used to measure HbA1c levels (VARIANT II Haemoglobin Testing System; Bio-Rad Laboratories), which is the gold standard method for monitoring glucose control in diabetes patients as recommended by the American Diabetes Association (ADA) [12]. The serum CP levels were determined using a chemiluminescent method and an Adivia Centaur XP immunoassay system (Siemens, Germany). The inter-assay and intra-assay variation coefficients of the CP test were 3.7-4.1% and 1.0-3.3%, respectively.

Continuous Glucose Monitoring
The blinded CGM system was used to generate dynamic glucose profiles (iPro2 with Enlite sensor, Medtronic MiniMed, Northridge, CA, USA). The CGM system's glucose sensor (MMT-7008A) was implanted on the lateral upper arm and removed after one week, generating a maximum daily record of 288 continuous sensor glucose measurements. Selfmonitoring of blood glucose (SMBG) was required for the participants in order to calibrate CGMs at least four times/day. Then, the CGM data were exported and subjected to quality assessment. When at least 70% of the CGM data (5 valid days, equivalent to 1440 glucose readings) was available, the CGM parameters were calculated using M-Smart software (CareLink iPro) provided by Medtronic.

Sample Preparation and Metabolomic Analysis
In positive and negative ESI modes, serum samples were analyzed using Liquid Chromatography Mass Spectrometer (LC-MS). The extraction of metabolites, instrument settings, peak intensities of metabolites, and differential regulation of metabolites between groups were determined as previously described [13,14]. First, the sample was thawed at 4 • C, vortexed for 1 min, and mixed evenly. Second, an appropriate amount of the sample was transferred into a 2 mL centrifuge tube. Third, 400 µL of methanol (stored at −20 • C) was added and vortexed for 1 min. Then, the supernatant was collected by centrifugation of the sample at 12,000 rpm, 4 • C for 10 min; it was further concentrated and dried. Finally, 150 µL of 2-chloro-l-phenylalanine (4 ppm) solution prepared with 80% methanol water (stored at 4 • C) was added to resolubilize the sample, and the supernatant was filtered through a 0.22 µm membrane and transferred into the detection bottle for LC-MS detection.

Statistical Analysis
Independent sample t-test, Mann-Whitney U test, and χ 2 test were used to compare the basic characteristics between GV-H and GV-L groups for normally distributed, skeweddistribution, and categorical variables, respectively.
The ropls R package from the Bioconductor repository was utilized for all multivariate data analyses and modeling [16]. Data were mean centered using scaling. Models were constructed using principal component analysis (PCA), partial least-square discriminant analysis (PLS-DA), and orthogonal partial least-square discriminant analysis (OPLS-DA). The VIP derived by OPLS-DA was used to screen out differential metabolites between the GV-H and GV-L groups. In addition, the metabolites identified by the OPLS-DA model were confirmed at a univariate level using the Wilcoxon-Mann-Whitney test (p < 0.05). The fold change (FC) was determined according to the relative quantification reflected by the peak intensity of the metabolites between the two groups (GV-H vs. GV-L). Finally, those with p < 0.05 and VIP > 1 were considered to be statistically significant metabolites. MetaboAnalyst, which combines the outcomes of potent pathway enrichment analysis and pathway topology analysis, was used to conduct pathway analysis on various metabolites. To adjust for confounding factors, a multiple regression analysis was conducted.
Spearman's correlation analysis was performed to evaluate the correlation between final selected differential metabolites and glycemic parameters. Multiple linear regression analysis was utilized to explore the relationship of glucose CV with age, BMI, duration of diabetes, FBG, HbA1c, daily insulin dosage, FCP, 2hCP, and final selected potential biomarkers by employing a stepwise procedure, which has been shown to be efficient in picking independent variables that are truly useful for predicting CV since it automatically eliminates the collinearity overstatement. The glucose CV was regarded as the dependent variable, while the other factors were regarded as independent variables. Receiver operating characteristics (ROC) curve analysis was used to determine the predictive ability of differential metabolites for GV and their performance as biomarkers. A two-tailed test was performed, and p < 0.05 was considered statistically significant. All statistical analyses were carried out using SPSS 26.0 software (IBM Corp., Armonk, NY, USA).

Comparison of Basic Characteristics and CGM Parameters of Study Subjects
A total of 33 patients with T1D were divided into GV-H (glucose CV ≥ 36%, n = 17) and GV-L (glucose CV < 36%, n = 16) groups based on the CV of glucose obtained from CGM ( Table 1). The mean glucose CV in the GV-H group was 46.3 ± 5.7% and 26.8 ± 3.7% in the GV-L group, respectively. No significant differences were observed in the age, gender, BMI, diabetes duration, daily insulin dosage, FBG, 2hBG, HbA1c, FCP, 2hCP, or lipid profiles between the two groups (all p > 0.05). Moreover, patients in the GV-H group also had a significantly higher standard deviation (SD) of glucose, mean amplitude of glucose excursions (MAGE), and low blood glucose index (LBGI) values than those in the GV-L group in addition to glucose CV (all p < 0.001).

Models Analysis in the Exploratory Set
A total of 5597 metabolites were identified (positive mode: 3012; negative mode: 2585). The score plots of the PCA (Figure 1a), PLS-DA (Figure 1b), and OPLS-DA ( Figure 1c) models were drawn for serum samples from the GV-H and GV-L groups in the exploratory set. The samples from the GV-H and GV-L groups were separated, indicating that they had significantly different metabolic profiles. The preliminary differential metabolites screening identified 569 differential metabolites from the HMDB and KEGG databases. These metabolites mainly included carboxylic acid and derivatives (20.21%), lipids (20.04%), benzene and substituted derivatives (11.78%), organic compounds (11.42%), purines, pyrimidines and their derivatives (6.68%), a few of the detailed classification unidentified metabolites (3.34%), and other metabolites ( Figure 1d).

Models Analysis in the Exploratory Set
A total of 5597 metabolites were identified (positive mode: 3012; negative mode: 2585). The score plots of the PCA (Figure 1a), PLS-DA (Figure 1b), and OPLS-DA ( Figure  1c) models were drawn for serum samples from the GV-H and GV-L groups in the exploratory set. The samples from the GV-H and GV-L groups were separated, indicating that they had significantly different metabolic profiles. The preliminary differential metabolites screening identified 569 differential metabolites from the HMDB and KEGG databases. These metabolites mainly included carboxylic acid and derivatives (20.21%), lipids (20.04%), benzene and substituted derivatives (11.78%), organic compounds (11.42%), purines, pyrimidines and their derivatives (6.68%), a few of the detailed classification unidentified metabolites (3.34%), and other metabolites (Figure 1d). Figure 1. Score plots of the PCA/PLS-DA/OPLS-DA models for discriminating the GV-H group from the GV-L group and pie chart of preliminary differential metabolites screening results in the exploratory set. Samples of these two groups were distinctly separated, indicating that they had markedly different metabolic characteristics (a-c). Pie chart (d) of differentially metabolomics showed that the 569 metabolites mainly involved carboxylic acids and derivatives (20.21%), lipids (20.04%), benzene and substituted derivatives (11.78%), organic compounds (11.42%), purines, pyrimidines, and their derivatives (6.68%), a few of the detailed classification unidentified metabolites (3.34%), and other metabolites.

Pathway Analysis in the Exploratory Set and Candidate GV Biomarkers
A total of 569 differential metabolites identified between the GV-H and GV-L groups were enriched in 82 KEGG metabolic pathways. The top twenty pathways altered between the GV-H and GV-L groups were picked according to the calculated −log10(p-value) value as displayed in Figure 2a. Additionally, the results indicated that the top seven pathways were altered significantly (p < 0.05, impact > 0.01): linoleic acid metabolism, aminoacyl-tRNA biosynthesis, ATP-binding cassette transporters, taurine and hypotaurine metabolism, phenylalanine metabolism, cysteine and methionine metabolism, and alanine, aspartate, and glutamate metabolism (Supplementary Table S1). A total of 25 metabolites were enriched in these seven pathways and were thus considered potential GV biomarkers (Table 2, Figure 2b). (a measure of the intensity and explanatory ability of the effect of different metabolite accumulation on the classification and discrimination of samples in each group; VIP ≥ 1 is a common differential metabolite screening criteria); Trend, FC > 1 indicates that the metabolite is down-regulated in the GV-H group, FC < 1 indicates that the metabolite is increased in the GV-H group compared with the GV-L group in the exploratory set. 9,10-DHOME, (12Z)-9,10-Dihydroxyoctadec-12-enoic acid; 12,13-DHOME, (9Z)-12,13-Dihydroxyoctadec-9-enoic acid.
(a) (b) Figure 2. Disturbed metabolic pathways and heatmap of preliminarily selected 25 differential metabolites between the GV-H and GV-L groups in the exploratory set. Bubble diagram of differential metabolic pathways between the GV-H and GV-L groups (a). Each point represents a pathway, the abscissa is the Impact value, and the ordinate is the enriched pathway. The dots indicate the number of metabolic molecules corresponding to the pathway. The color is related to the P value, the redder the color, the smaller the P value. Heatmap of preliminarily selected 25 differential metabolites (b). Each row represents a metabolite, and each column depicts a subject. Group A is the GV-L group and group B is the GV-H group; the peak intensity of a metabolite is shown by different colors, namely, the redder the color, the higher the peak intensity.

Candidate GV Biomarkers Validation
Another 21 T1D patients underwent metabolomics analysis to further validate the stability of these 25 differential metabolites (Supplementary Table S2). To reduce bias caused by potential confounders, these 21 subjects were divided into 11 with high GV (mean glucose CV = 46.2%) and 10 matched patients with low GV (mean glucose CV = 25.3%). Although phosphatidylcholine, riboflavin, 9,10-DHOME, cysteine-S-sulfate, Lcysteine, hydrocinnamic acid, N-acetyl-L-aspartate, and mannitol were significantly different in the GV-H group compared to the GV-L group in this validation set, their changing trends were opposite to those observed in the exploration set. On the other hand, spermidine, trehalose, and L-methionine were altered similarly to those in the exploratory set (Supplementary Table S3). Thus, we finally identified these three differential metabolites mentioned above as potential GV biomarkers.

Performance of Final Selected Biomarkers for Predicting GV
In the exploratory set, spermidine outperformed in distinguishing two groups with an area under the curve (AUC) of 0.879 (95% confidence interval (CI): 0.721-1.000) and sensitivity and specificity of 100.0% and 87.5%, respectively. Trehalose had an AUC of 0.886 (95% CI: 0.772-1.000, sensitivity: 81.3%, specificity: 82.4%), and L-methionine had an AUC of 0.746 (95% CI: 0.577-0.915, sensitivity: 56.3%, specificity: 88.2%), respectively. Furthermore, the AUC of the above three metabolites combined to differentiate the GV-H and Figure 2. Disturbed metabolic pathways and heatmap of preliminarily selected 25 differential metabolites between the GV-H and GV-L groups in the exploratory set. Bubble diagram of differential metabolic pathways between the GV-H and GV-L groups (a). Each point represents a pathway, the abscissa is the Impact value, and the ordinate is the enriched pathway. The dots indicate the number of metabolic molecules corresponding to the pathway. The color is related to the p value, the redder the color, the smaller the p value. Heatmap of preliminarily selected 25 differential metabolites (b). Each row represents a metabolite, and each column depicts a subject. Group A is the GV-L group and group B is the GV-H group; the peak intensity of a metabolite is shown by different colors, namely, the redder the color, the higher the peak intensity. Note: FC, fold change (the peak intensity ratio of target metabolite in the GV-L and GV-H groups); p value, p value of hypergeometric distribution test (the smaller p value, the more significant the effect of detected differential metabolites on this pathway); VIP, variable importance in projection (a measure of the intensity and explanatory ability of the effect of different metabolite accumulation on the classification and discrimination of samples in each group; VIP ≥ 1 is a common differential metabolite screening criteria); Trend, FC > 1 indicates that the metabolite is down-regulated in the GV-H group, FC < 1 indicates that the metabolite is increased in the GV-H group compared with the GV-L group in the exploratory set. 9,10-DHOME, (12Z)-9,10-Dihydroxyoctadec-12-enoic acid; 12,13-DHOME, (9Z)-12,13-Dihydroxyoctadec-9-enoic acid.

Candidate GV Biomarkers Validation
Another 21 T1D patients underwent metabolomics analysis to further validate the stability of these 25 differential metabolites (Supplementary Table S2). To reduce bias caused by potential confounders, these 21 subjects were divided into 11 with high GV (mean glucose CV = 46.2%) and 10 matched patients with low GV (mean glucose CV = 25.3%). Although phosphatidylcholine, riboflavin, 9,10-DHOME, cysteine-S-sulfate, L-cysteine, hydrocinnamic acid, N-acetyl-L-aspartate, and mannitol were significantly different in the GV-H group compared to the GV-L group in this validation set, their changing trends were opposite to those observed in the exploration set. On the other hand, spermidine, trehalose, and L-methionine were altered similarly to those in the exploratory set (Supplementary  Table S3). Thus, we finally identified these three differential metabolites mentioned above as potential GV biomarkers.

Correlation between Selected Biomarkers and Glycemic Parameters for all Patients
L-methionine was significantly inversely related to HbA1c (r = −0.427, P = 0.001) and FBG (r = −0.329, P = 0.017). In terms of CGM parameters, L-methionine and trehalose were inversely related to low blood glucose index (LBGI) (both P < 0.05). On the other hand, spermidine was significantly positively correlated, whereas L-methionine was negatively correlated with SD, MAGE, glucose CV, and LBGI (all P < 0.01) (Table 3, Figure 4). Table 3. Correlation of selected biomarkers with glycemic parameters in all subjects (r).

Correlation between Selected Biomarkers and Glycemic Parameters for All Patients
L-methionine was significantly inversely related to HbA1c (r = −0.427, p = 0.001) and FBG (r = −0.329, p = 0.017). In terms of CGM parameters, L-methionine and trehalose were inversely related to low blood glucose index (LBGI) (both p < 0.05). On the other hand, spermidine was significantly positively correlated, whereas L-methionine was negatively correlated with SD, MAGE, glucose CV, and LBGI (all p < 0.01) (Table 3, Figure 4).

Predictors for GV by Multiple Linear Regression Analysis for All Patients
The correlation between glucose CV and age, BMI, diabetes duration, daily insulin dosage, FBG, HbA1c, FCP, 2hCP, spermidine, L-methionine, and trehalose was analyzed

Discussion
T1D is an organ-specific autoimmune disease in which pancreatic β-cells suffer varying degrees of immune damage, resulting in absolute insulin deficiency and non-negligible GV [1]. With the increasing use of CGM, several studies have revealed that GV may be involved in the emergence of diabetic complications [17][18][19], although the mechanism is not yet understood. Moreover, GV in some T1D patients remained heterogeneous despite comparable clinical indicators, whether other factors are involved is yet unknown.
In the exploratory set, we identified 569 metabolites that differed between the GV-H and GV-L groups after the primary screening. These metabolites were enriched in 82 metabolic pathways, of which seven were identified as differential metabolic pathways. Subsequently, 25 different metabolites were found to be enriched in these seven pathways.

Discussion
T1D is an organ-specific autoimmune disease in which pancreatic β-cells suffer varying degrees of immune damage, resulting in absolute insulin deficiency and non-negligible GV [1]. With the increasing use of CGM, several studies have revealed that GV may be involved in the emergence of diabetic complications [17][18][19], although the mechanism is not yet understood. Moreover, GV in some T1D patients remained heterogeneous despite comparable clinical indicators, whether other factors are involved is yet unknown.
In the exploratory set, we identified 569 metabolites that differed between the GV-H and GV-L groups after the primary screening. These metabolites were enriched in 82 metabolic pathways, of which seven were identified as differential metabolic pathways. Subsequently, 25 different metabolites were found to be enriched in these seven pathways. After validation, three candidate GV biomarkers were identified: spermidine, L-methionine, and trehalose. Furthermore, ROC analysis revealed that spermidine, L-methionine, and trehalose adequately distinguished between the GV-H and GV-L groups in both the exploratory and validation sets. In addition to BMI and FCP, multiple linear regression analysis revealed that L-methionine and spermidine were independent predictors of glucose CV.
Spermidine (C 7 H 19 N 3 ), such as spermine and putrescine, are natural polyamine compounds [20]. L-arginine is used as a substrate to synthesize polyamines in mammalian cells via arginase in extrahepatic tissues. Polyamines are a class of compounds that contain two or more amino groups and are primarily synthesized from L-arginine and ornithine, with arginine decarboxylase and ornithine decarboxylase serving as key enzymes [21]. Polyamines have been shown to be effective antioxidants that protect essential cellular components in cell membranes, such as polyunsaturated fatty acids in the membrane, from oxidative damage over the last decade [22]. Higher amounts of polyamines have been documented in both exocrine and endocrine pancreatic cells, which may lead to endoplasmic reticulum stress, oxidative stress, inflammatory response, and autophagy [23]. Specifically, polyamines have been shown to play a pathogenic role in T1D progression because they are known to play a critical role in the functioning of β-cells [24] and immune cells [25]. The inhibition of polyamine biosynthesis significantly delays diabetes incidence in NOD mice [26]. Moreover, polyamines have a time-and concentration-dependent inhibitory effect on the protein phosphatase activity of insulin-secreting cells in diabetes [27]. In vivo, spermidine inhibited HbA1c and lipid peroxidation [28]. However, a recent study on the role of spermidine treatment in T1D pathogenesis found that daily oral spermidine treatment in NOD mice resulted in a higher diabetes incidence, organ-specific spermidine accumulation, elevated peripheral inflammation, and a lower proportion of suppressive Tregs [29]. Reportedly, spermidine is linked to diabetic complications. Spermidine levels in the aqueous humor of patients with proliferative diabetic retinopathy (PDR) can be 3-4 times higher than in healthy controls [30]. Furthermore, in insulin-treated diabetic patients, an increase in intra-erythrocytic spermidine content has been linked to both diabetic nephropathy and severe retinopathy [31]. It has also been discovered that children with T1D have higher polyamine oxidase activity, which could lead to increased ROS production and subsequent oxidative damage [32]. Similarly, spermidine oxidase activity, another polyamine catabolic enzyme, is significantly lower in T1D patients compared to non-diabetics [33]. This highlights the fact that decreased activity of polyamine catabolic enzymes in T1D patients may promote polyamine accumulation.
According to our findings, spermidine levels were higher in patients with high GV, and spermidine levels were significantly positively correlated with GV parameters. This phenomenon could be attributed to increased oxidative stress in patients with high GV [34] because oxidative stress caused by GV would be more severe than hyperglycemia alone [33,[35][36][37]. The stimulation of oxidative stress by GV is associated with an increase in ROS production, oxidative damage to DNA, and a decrease in superoxide dismutase activity [38]. The current study was the first to show that spermidine levels can be used to predict GV levels in T1D patients, implying that higher spermidine levels are associated with high GV. Thus, combining the current findings with the previously reported tendency of increased spermidine levels in patients with microvascular complications may provide an in-depth insight into the role of GV in the development of diabetic complications. The mechanism of elevated spermidine levels in T1D individuals with high GV is yet to be elucidated. Given that oxidative stress brought on by GV would be more severe than hyperglycemia alone [34][35][36][37][38], one explanation is that spermidine excess is a reactive change brought on by reactive oxygen species (ROS). GV-induced oxidative stress is related to an increase in ROS generation, oxidative DNA damage, and a reduction in superoxide dismutase function [39]. Reportedly, higher ROS generated spermidine, which protected the cell from oxidative damage [40,41]. In this context, changes in spermidine concentration could be a consequence, and may not be the cause, of high GV. So far, however, there are only assumptions regarding the precise mode of action and the impact on GV of different spermidine concentrations. To support or refute any of the foregoing hypotheses, additional in vitro and in vivo mechanistic research is required.
L-methionine (C 5 H 11 NO 2 S) is an essential sulfur-containing α-amino acid that cannot be produced in the body and is only available externally [42]. L-methionine has been shown to inhibit liver gluconeogenesis gene expression by promoting PGC1-acetylation [43]. Lmethionine and other methyl donors improve glucose tolerance and insulin sensitivity in mice offspring fed a high-fat diet [44]. Accumulating evidence indicated that L-methionine activates AMPK and SIRT1, a mechanism similar to metformin. Furthermore, L-methionine has been shown to improve the altered key one-carbon metabolite metabolism in T2D rats and diabetes-induced disturbances in glucose and lipid metabolism [45]. These findings suggested that increased methionine levels could be used to treat diseases associated with glucose and lipid metabolism disorders. Patients in the GV-H group had significantly lower serum L-methionine levels than those in the GV-L group in the current study, and L-methionine was inversely correlated with both HbA1c and GV parameters in correlation analysis; thus, L-methionine was an independent predictor of glucose CV, indicating its role as a biomarker of GV in T1D patients. The role of methionine in T1D pathogenesis and progression has also been indicated in several studies. In T1D rats, L-methionine also protected pancreatic β-cells by regulating FOXO1 expression [46]. According to the German BABYDIAB study, children who developed autoantibodies by the age of two had a 2-fold lower methionine content than those who developed autoantibodies later or autoantibody-negative controls [47]. Another 'Environmental Triggers for Type 1 Diabetes' (MIDIA) study found that methionine levels in T1D progressors were lower with time, according to longitudinal changes in plasma metabolic profiles [48]. Methionine deficits were also detected in the metabolomics profile of the pre-T1D mice [49]. Furthermore, there is evidence that low methionine levels are associated with increased oxidative stress. Significant T1D-dependent increases in circulating oxidation products and decreases in methionine and cysteine levels were detected, indicating increased oxidative stress [50]. Methionine is a component of the transsulfuration process, which leads to glutathione production, an important intracellular antioxidant. Prior studies have shown that insulindeprived T1D patients have lower rates of homocysteine-methionine remethylation and higher rates of transsulfuration compared to control subjects [51]. Accordingly, patients with poor glycemic control have lower glutathione pools and lower erythrocyte free cysteine level that can be synthesized from methionine [52]. Therefore, the decrease in circulating methionine may indicate an increased glutathione production. In the present study, patients in the GV-H group had significantly lower serum L-methionine levels than those in the GV-L group, and L-methionine was inversely correlated with HbA1c, FBG, and GV parameters in correlation analysis. Consequently, we speculate that decreases in circulating methionine may serve as a potential biomarker for T1D patients with poor glycemic control and high GV.
Trehalose (C 12 H 22 O 11 ) is a non-reduced disaccharide composed of two glucose molecules joined by a hemiacetal hydroxyl group. Despite the fact that trehalose is a non-reduced disaccharide, several in vitro and in vivo studies have confirmed its role as a natural antioxidant [53,54]. Trehalose has antioxidative, anti-inflammatory, and enhanced autophagy functions that can inhibit oxidative stress, inflammation, and autophagy-related diseases, such as diabetes [55,56] and atherosclerosis [57]. Specifically, trehalose's anti-diabetic effect may be linked to reduced oxidative stress and improved islet function [58]. Furthermore, trehalose induces CD8 + Treg cells in mice and preserves insulin concentration, a critical molecule in inhibiting STZ-induced T1D development in mice. Trehalose supplementation completely reverses the mild hyperglycemia (<19.4 mmol/L) in diabetic NOD mice and might also increase CD8 + Treg cells in mice that did not respond to treatment. These findings suggested that trehalose could be used to treat T1D when pancreatic β-cell regeneration is possible [59]. However, metabolomics studies revealed that trehalose might be a risk factor for the development of T2D [60,61]. Another metabonomic study on the risk factors for DR in T2D patients showed that low serum trehalose concentrations were associated with DR progression [62]. According to the current findings, serum trehalose levels are higher in T1D patients with low GV. Additionally, the present study was the first to use metabolomics to identify the link between trehalose and GV. It can be seen that uncertainty exists regarding the association between elevated trehalose levels and high GV in T1D patients. The levels of trehalose reflected by the peak intensity in the present study were very low when compared to spermidine and L-methionine. Moreover, it is worth noting that in both the exploratory and validation sets, the FC values of spermidine and L-methionine were relatively consistent. The FC of trehalose in the validation set, however, was less than half that of the FC of trehalose in the exploration set, indicating that the difference of trehalose between the GV-H and GV-L groups may not as reliable as that of spermidine and L-methionine. Therefore, additional in vitro and in vivo mechanistic studies are required to clarify the relationship between high trehalose levels and high GV.
The present study has some limitations. First, the sample size was small. Although we matched the GV-H and GV-L groups according to age, gender, diabetes duration, BMI, daily insulin dosage, FBG, HbA1c, and CP levels to minimize the confounding factors, the results may be underpowered for the detection of certain biomarkers specific to GV. Second, due to the cross-sectional study design, we could not elucidate the cause-and-effect correlation in this study. Third, future studies are required to test the performance and reliability of selected biomarkers in patients with different levels of GV. One other potential limitation of our study is that we were unable to assess the impact of dietary intake on the circulatory metabolome of T1D patients, which may have contributed to the differences in levels of the present identified metabolites. Future investigation into diet-metabolome interactions will be required to clarify the impact of dietary intake on glycemic control and GV levels of patients with T1D.

Conclusions
In conclusion, the GV-H and GV-L groups exhibited different metabolic perturbations. Herein, we reported for the first time that increased spermidine and decreased L-methionine and trehalose may contribute to high GV in T1D patients. BMI, FCP, spermidine, and Lmethionine were independent predictors of glucose CV. Thus, future studies should focus on determining and validating whether the identified metabolites are related to an increased risk of micro-and macrovascular complications in T1D patients in a prospective, randomly selected population. In conclusion, the current study emphasizes the various metabolic disturbances among T1D patients with different GV levels. We discovered that serum spermidine, L-methionine, and trehalose may impact glycemic control; particularly, high spermidine and low L-methionine levels were associated with high GV. L-methionine was shown to be more closely associated with poor glycemic control and high GV. Differences in trehalose were not as reliable as those in spermidine and L-methionine because of the relatively low amounts of trehalose and the inconsistent fold change sizes in the exploratory and validation sets. These alterations, however, may not be specifically related to GV since dietary intake may also have a partial influence on the metabolomics profiles. Our results also suggest that spermidine and L-methionine, in addition to BMI and FCP, may be useful predictors of glucose CV. Overall, these observations suggest that metabolomic disturbances impact the GV of T1D. Additional in vitro and in vivo mechanistic studies are required to elucidate the relationship between spermidine and L-methionine levels and GV in a cohort of T1D patients with different geographical and nutritional backgrounds.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/nu15030518/s1, Table S1: Identification of differential metabolic pathways in GV-H group and GV-L group in the exploratory set; Table S2: Basic characteristics and CGM parameters in GV-H and GV-L groups of the validation set; Table S3: Performance of candidate metabolic markers in the validation set. Table S4: Basic characteristics and CGM parameters of patients enrolled and not enrolled. Figure S1: Flow chart of the comprehensive analysis of metabolomics profiles of patients with different glycemic variability.
Author Contributions: Study design, writing-review and editing: Z.Z. and L.Y.; original draft preparation and data analysis: L.Z.; data collection: K.G., Q.T., J.Y., Z.D., Q.Z. and X.L. All authors have read and agreed to the published version of the manuscript. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study. Data Availability Statement: All relevant data are within the manuscript and its Supporting Information files.