Identification of individuals with decreased functional β-cell mass is essential for the prevention of diabetes. However, in vivo detection of early asymptomatic β-cell defect remains unsuccessful. Metabolomics has emerged as a powerful tool in providing readouts of early disease states before clinical manifestation. We aimed at identifying novel plasma biomarkers for loss of functional β-cell mass in the asymptomatic prediabetes stage. Nontargeted and targeted metabolomics were applied in both lean β-Phb2−/− (β-cell-specific prohibitin-2 knockout) mice and obese db/db (leptin receptor mutant) mice, two distinct mouse models requiring neither chemical nor dietary treatments to induce spontaneous decline of functional β-cell mass promoting progressive diabetes development. Nontargeted metabolomics on β-Phb2−/− mice identified 48 and 82 significantly affected metabolites in liver and plasma, respectively. Machine learning analysis pointed to deoxyhexose sugars consistently reduced at the asymptomatic prediabetes stage, including in db/db mice, showing strong correlation with the gradual loss of β-cells. Further targeted metabolomics by gas chromatography–mass spectrometry uncovered the identity of the deoxyhexose, with 1,5-anhydroglucitol displaying the most substantial changes. In conclusion, this study identified 1,5-anhydroglucitol as associated with the loss of functional β-cell mass and uncovered metabolic similarities between liver and plasma, providing insights into the systemic effects caused by early decline in β-cells.

Both type 1 and type 2 diabetes (T2D) are chronic diseases primarily characterized by a β-cell defect. While the former is an autoimmune disease resulting in the inexorable destruction of β-cells (1), the latter is usually associated with obesity-induced insulin resistance, eventually leading to β-cell failure that prompts diabetes (2,3). Indeed, insufficient insulin secretion has emerged as the determining factor for the onset of T2D (4,5). Although loss of β-cell function and mass both contribute to impaired insulin secretion (6,7), these two parameters are not well correlated, and their relative contribution to T2D pathogenesis remains a subject of debate (8,9). Identification of robust and sensitive in vivo biomarkers that could reflect the integrated assessment of β-cell function and mass, i.e., the functional β-cell mass, would be essential for the identification of high-risk yet asymptomatic individuals for the prevention of diabetes.

Metabolomics is a powerful method to measure subtle biochemical changes caused by underlying pathologies (10,11). The main metabolites consistently appearing in human T2D studies include branched-chain amino acids (BCAAs) (12,13), aromatic amino acids (13), fructose (14,15), mannose (14,16), α-hydroxybutyrate (17,18), and phospholipids (1820). Most of these identified T2D-related metabolites are associated with insulin resistance and obesity (21,22) rather than the direct alteration of β-cells. This is partially due to the fact that the clinical practice for diabetes detection, such as fasting glucose and glucose tolerance tests, cannot efficiently discriminate the respective contributions of insulin resistance versus insulin secretion (23,24). Therefore, we still lack biomarkers reflecting the loss of functional β-cell mass as the triggering event.

The main obstacle for the identification of biomarkers that would be specific for early β-cell defects is the scarce availability of pancreas samples from subjects with prediabetes. In essence, one cannot study isolated human pancreas from an individual who would later eventually develop T2D; likewise, pancreatic tissues are available only upon autopsy or invasive surgery (25). Animal models are useful surrogates but typically require special diets or drug injections to induce diabetes, which may generate confounding factors. To address these issues, we used two mouse models: β-cell-specific prohibitin-2 knockout (β-Phb2−/−) mice (26) and leptin receptor–deficient (db/db) mice (27). Both mouse models exhibit progressive β-cell failure and develop diabetes, although the genetic cause of the disease differs. Deletion of prohibitin-2 in the β-cells causes mitochondrial alterations and β-cell dysfunction, then apoptosis and a progressive decline in β-cell mass starting at 4 weeks of age resulting in the onset of diabetes in β-Phb2−/− mice at 6 weeks of age (26). The db/db mice carry a single-gene spontaneous mutation in the leptin receptor that leads to hyperphagia and obesity (28); diabetes appears at the age of ∼8 weeks (29) when the decline in functional β-cells results in the failure to compensate for insulin resistance (30). In both mouse models, the critical event triggering diabetes, subsequently uncovered by hyperglycemia, is the loss of a critical mass of functional β-cells, similar to what occurs in human subjects (5,31). Using nontargeted and targeted metabolomics, we profiled the liver and plasma metabolome of β-Phb2−/− and db/db mice from the earliest postweaning age (4 weeks) to diabetes onset and later. Along with the assessment of β-cell mass, simultaneous measurements of hundreds of metabolites (32) led to the identification of common changing metabolites. Ultimately, we found a shared metabolite among the mouse models; its levels linearly correlate with the decline of functional β-cell mass that already occurs in the asymptomatic prediabetic state.

Animals

All animal experiments were conducted at the University of Geneva Medical Centre with the approval of the animal care and experimentation authorities of the Canton of Geneva (#GE/128/15). Mice were maintained on a 12-h dark/12-h light cycle with water and food ad libitum and genotyped using primers listed in Supplementary Table 1. Since the phenotype of β-Phb2−/− mice is similar between males and females (26), only males were used for experiments; the age of the mice is specified for each experiment. All β-Phb2−/− and Phb2fl/fl mice were generated as previously described (26) and maintained on a mixed genetic background (C57BL/6J × 129/Sv) to avoid inbred strain-specific phenotypes. BKS.Cg-Dock7m+/+Leprdb/J mice (db/db and heterozygous db/+ mice as controls) were purchased from Charles River Laboratories Italia, Calco, Italy.

Plasma and Tissue Collection

On the day of sacrifice, food was removed from cages at 8:30 a.m., and mice were sacrificed from 11:30 a.m. until 12:00 p.m. In mice, this 3-h fasting period corresponds to a physiological overnight fast in human subjects, i.e., before total exhaustion of hepatic glycogen stores (33), while starving mechanisms are not yet operating (34). Body weights and glucose levels were monitored just before sacrifice. Blood glucose values were measured using an Accu-Chek Aviva glucometer (Roche Diagnostics). Blood was collected into EDTA-coated tubes (no. 20.1341; Sarstedt, Inc.) via retro-orbital bleeding and centrifuged at 2,000 rpm at 4°C in order to separate plasma. C-peptide was measured with a commercial kit (Mouse C-Peptide ELISA Kit, catalog no. 90050; Crystal Chem). We calculated the HOMA of insulin resistance (HOMA-IR) index (35) as follows: insulin (µU/mL) × fasting glycemia (mmol/L)/22.5. Tissues of interest were collected, weighed, and snap frozen in liquid nitrogen and stored at −80°C. Pancreas was collected for fixation and further immunohistochemistry.

Immunostaining and β-Cell Mass Quantification

Upon sacrifice, pancreas was fixed for 2 h in 4% paraformaldehyde and embedded in paraffin. Sections of 5 µm were incubated overnight with primary antibody and then with secondary antibody for 1 h (Supplementary Table 2). Images were captured by confocal microscopy (LSM 800; Zeiss). For assessment of β-cell mass, 5-µm sections at an interval of 150 µm throughout the pancreas were stained for insulin with the primary antibody and horseradish peroxidase–conjugated secondary antibody and then visualized by DAB Substrate Chromogen System (K3468; Dako). Sections were scanned by using an Axio Scan.Z1 widefield slide scanner (Zeiss), and brown areas were quantified using Definiens software, adjusted to total pancreas weight.

Insulin Secretion on Isolated Islets

Pancreatic islets were isolated through collagenase digestion and cultured overnight in complete RPMI 1640 medium, as described previously (36). For secretion assays, islets were preincubated with Krebs-Ringer bicarbonate HEPES buffer supplemented with 2.8 mmol/L glucose and 0.1% BSA for 1 h. Subsequently, batches of 10 islets were hand-picked and incubated with basal glucose (2.8 mmol/L) and stimulatory glucose (22.2 mmol/L) at 37°C for 1 h. At the end of the incubation period, supernatants were collected and islets were resuspended in acid ethanol for determination of insulin contents. Insulin in supernatants and extracted islets was measured using radioimmunoassay (Millipore).

Metabolome Extraction

The metabolome extractions were performed from 50–100 mg of frozen liver. The samples were initially homogenized by TissueLyser II (Qiagen) in 1 mL cold 70% ethanol. Extraction was continued with addition of 7 mL hot (75°C) 70% ethanol for 2 min with thorough mixing every 30 s, followed by removal of debris by centrifugation. Extracts were stored at −20°C until mass spectrometry (MS) analysis. Plasma was extracted by adding 90 μL of 80% methanol to a 10-μL volume of plasma, followed by thorough vortex mixing. After incubation at 4°C for 1 h, centrifugation was done at maximum speed for 10 min and supernatant was used for MS measurements.

Nontargeted Metabolomics

Untargeted metabolomics was performed by flow injection analysis (FIA) on an Agilent 6550 iFunnel Q-TOF LC/MS instrument as described previously (32). Briefly, profile spectra were recorded in negative ionization from m/z 50 to 1,000 at 4-GHz high-resolution mode. Ion annotation was based on accurate masses, using a tolerance of 0.001 amu and organism code “mmu” in the KEGG database, accounting for –H+ and F ions, sodium and potassium adducts, and heavy isotopes. Further data analysis involved calculating log2 fold changes between different groups (e.g., knockout vs. control at week 4), with statistical significance calculated by two-sample Student t test, additionally corrected for multiple testing and expressed as q values (37). We analyzed data using MATLAB software (The MathWorks).

For targeted metabolomics by liquid chromatography–MS, polar metabolome extracts were dried and resuspended in deionized water. In order to adjust for differences in tissue sample weight, dried samples were resuspended in 10 µL deionized water per 1 µg tissue. Centrifuged cold supernatant aliquots of 10 µL were injected into an Acquity UPLC System (Waters Corporation, Milford, MA) with an ACQUITY T3 column (Waters Corporation) coupled to a Thermo TSQ Quantum Ultra triple quadrupole instrument (Thermo Fisher Scientific) with negative-mode electrospray ionization. A gradient of two mobile phases—10 mmol/L tributylamine, 15 mmol/L acetic acid, and 5% (v/v) methanol (phase A) and 2-propanol (phase B)—was used for separation of compounds (38). Peak integration of mass spectra was performed by in-house software (B. Begemann and N. Zamboni, unpublished observations).

Metabolomics Targeted to Deoxysugar Separation

Targeted metabolite profiling to separate deoxysugars was performed as detailed previously (39,40) by gas chromatography (GC) coupled to electron impact ionization/time-of-flight (TOF) mass spectrometry using an Agilent 6890N24 gas chromatograph (Agilent Technologies, Böblingen, Germany) with split and splitless injection onto a VF-5ms capillary column (30 m length, 0.25 mm inner diameter, and 0.25 μm film thickness) (Agilent Technologies, Waldbronn, Germany), which was connected to a Pegasus III TOF mass spectrometer (LECO Instrumente GmbH, Mönchengladbach, Germany). We performed enhanced targeted metabolite profiling using GC–atmospheric pressure ionization–quadrupole TOF MS (Bruker Daltonik GmbH, Bremen, Germany) as previously described (41).

Aliquots of 50 μL from the polar metabolite fractions (metabolome extracts; see above) were dried by vacuum concentration and stored dry under inert gas at −20°C until further processing. Metabolites were methoxyaminated and trimethylsilylated manually prior to GC/electron ionization TOF MS analysis (39,40). Retention indices were calibrated by addition of a C10, C12, C15, C18, C19, C22, C28, C32, and C36 n-alkane mixture to each sample (42).

GC/electron ionization TOF MS chromatograms were acquired, visually controlled, baseline corrected, and exported in NetCDF file format by using ChromaTOF software (version 4.22; LECO, St. Joseph, MI). GC-MS data processing into a standardized numerical data matrix and compound identification were performed using the TagFinder software (43). Compounds were identified by mass spectral and retention time index matching to the reference collection of the Golm Metabolome Database (http://gmd.mpimp-golm.mpg.de/) (44) and to the mass spectra of the NIST17 database (https://www.nist.gov/srd/shop). Guidelines for manually supervised metabolite identification were the presence of at least three specific mass fragments per compound and a retention index deviation <1.0% (42). Authenticated reference compounds for identification were obtained—namely, 1,5-anhydro-d-glucitol (A7165, CAS 154-58-5), D-(+)-fucose (F8150, CAS 3615-37-0), and L-(−)-fucose (F2252, CAS 2438-80-4)—from Sigma Aldrich GmbH (Steinheim, Germany). We did not distinguish between D and L stereoisomers. Identification of 1,5-anhydro-d-glucitol was supported by earlier reports (45) and by exact masses (±5 mDa) of the molecular ion (M+), proton adducts (M+H+), and M-CH3+ fragments that were observed on GC/atmospheric pressure chemical ionization quadrupole TOF MS.

All mass features of an experiment were normalized by sample fresh weight or volume (plasma) and internal standard and subsequently scaled (percentage) in order to visualize relative concentration changes.

Machine Learning Algorithm

The machine learning method to classify the disease state consists of a composition of a preprocessing step and a classification step. For the preprocessing step, we tested the following algorithms: principal component analysis (PCA), L1-regularized logistic regression, and correlation thresholding, which consists of ranking the metabolites according to their Pearson sample correlation coefficient and selecting those with the highest correlation coefficient (the number of chosen features is a parameter of the algorithm, coded in Python). For the classification step, we tested linear discriminant analysis, L1-regularized logistic regression, gradient boosting, and random forests (46). After testing all the possible combinations of preprocessing and classification algorithms listed above, we selected the one yielding the best score on a leave-20-out cross-validation procedure repeated 100 times. Overall, the percentage of sampling data into validation sets was 18% and 28% of the total data set for β-Phb2−/− and db/db mice, respectively. Each time, the validation set was resampled exclusively from the set of metabolomics profiles corresponding to the prediabetic stage. The whole machine learning pipeline was coded in Python, making extensive use of the “sklearn” package (47). The latest version of the program used for this analysis is available for download (https://github.com/agazzian/ml_project).

We consistently obtained the best cross-validation scores by combining a correlation thresholding method for the preprocessing step with a random forests algorithm for the classification step. The metabolites were finally ranked according to their importance in the classification process (attribute feature_importances_ of the sklearn.RandomForestClassifier estimator). It should be noted that new random forest models were generated for each classification run, since the goal was not to find a universal classifier but rather to perform feature extraction, i.e., to investigate the classifying power of the selected metabolites.

Statistical Analysis

Data presented as dot plots are shown as mean ± SEM. The box plots are presented as median with 25th/75th percentiles (boxes) and 10th/90th percentiles (bars). Statistical parameters (numbers and P values) are included in each figure legend, except for number of mice used for metabolomics analyses, which is listed in Supplementary Table 3. For statistical analysis, the Mann-Whitney U test was used to compare two groups, unless described otherwise. A P value <0.05 was considered significant.

Data and Resource Availability

The data sets generated and analyzed during this study are available from the corresponding author upon reasonable request.

Phenotypic Characterization of β-Phb2−/− and db/db Mice

The β-Phb2−/− mice exhibit spontaneous progression from euglycemia to diabetes over a 2-week period, independently of any chemical or dietary treatments (26). The primary cause of diabetes, efficiently treated by insulin therapy (26), is β-cell failure rooted in mitochondrial dysfunction and accompanied by the loss of the β-cell mass. We further assessed the β-Phb2−/− mouse model through phenotyping at the prediabetic stage (4 and 5 weeks of age) and at the diabetic stage (6 and 10 weeks of age) (Fig. 1). At the ages of 4 and 5 weeks, β-Phb2−/− mice had normal levels of blood glucose whereas at 6 weeks they became hyperglycemic (glucose >11.1 mmol/L), and by the age of 10 weeks they were severely diabetic. The body weights of β-Phb2−/− mice were similar to those of the controls at the prediabetic stage. Once β-Phb2−/− mice developed diabetes, their body weights became lower than those of the controls (Fig. 1B). Adipose tissue was comparable between β-Phb2−/− mice and the controls until diabetes onset and then remarkably reduced when β-Phb2−/− mice developed diabetes (Fig. 1C).

Consistent with previous characterization of the widely studied mouse model for T2D (2830), we observed that db/db mice developed diabetes at the age of 8 weeks (Fig. 1D). In contrast to β-Phb2−/− mice, db/db mice had higher body weights than their controls at the prediabetic stage (4 and 6 weeks of age) and at diabetes onset (8 weeks of age) (Fig. 1E). The db/db mice had much more adipose tissue than their controls at all examined time points (Fig. 1F).

Islet Morphology and β-Cell Mass in β-Phb2−/− Mice

Before the appearance of symptoms, the development from prediabetes to diabetes was accompanied by marked changes in the islet morphology of β-Phb2−/− mice. There was a gradual decrease of insulin-positive cells and intrusion of glucagon-positive cells within the core of the islets (Fig. 2A). Notwithstanding β-Phb2−/− mice exhibiting euglycemia at 4 and 5 weeks of age (Fig. 1A), quantitative analysis revealed a clear reduction (58–64%) of their β-cell mass when compared with that in controls (Fig. 2B and C). This correlated with severely impaired glucose-stimulated insulin secretion tested in islets isolated from 4-week-old β-Phb2−/− mice (26). In mice aged 6 weeks, the observed 68% reduction of β-cell mass was then associated with hyperglycemia (Fig. 1A). Neither circulating C-peptide (Fig. 2D) nor HOMA-IR (Fig. 2E) could reflect the reduction of β-cells before the appearance of hyperglycemia. Of note, C-peptide was undetectable in β-Phb2−/− mice at 10 weeks of age (data not shown). Therefore, β-Phb2−/− mice present asymptomatic early-onset reduction of β-cell mass before the appearance of hyperglycemia at age 6 weeks. This indicates a threshold of approximately one-third of the β-cell mass for maintaining euglycemia in such nonobese animals.

Islet Morphology and β-Cell Mass in db/db Mice

As opposed to β-Phb2−/− mice, db/db mice preserved their insulin-positive cells during diabetes development (Fig. 3A). The β-cell mass of db/db mice doubled from the age of 4 weeks to 8 weeks and was significantly higher than that in their controls at all examined time points (Fig. 3B and C). Consistent with the β-cell mass, insulin content per islet increased in db/db mice compared with db/+ controls at 4 and 8 weeks of age (35.0 ± 3.7 vs. 18.5 ± 2.4 and 48.4 ± 2.5 vs. 14.6 ± 3.8 ng/islet, respectively), along with circulating C-peptide (Fig. 3D). In parallel, the insulin secretory responses of isolated islets from db/db mice dramatically deteriorated as diabetes developed (Supplementary Table 4). At the age of 4 weeks, db/db mice exhibited glucose-stimulated insulin secretion similar to that of the controls. However, upon appearance of hyperglycemia at 8 weeks, db/db mice presented a sharp decline of β-cell function as reflected by an ∼80% reduction of glucose-stimulated insulin secretion, compared with their controls. This implies that, despite the observed increase in insulin-positive cells, the functional β-cell mass of db/db mice actually decreased during diabetes progression, consistent with previous reports (30). The HOMA-IR index did not correlate with the β-cell mass (Fig. 3E). Taken together, these data show that β-Phb2−/− and db/db mice share a progressive decline of functional β-cell mass at a prediabetic asymptomatic stage.

Metabolite Changes in Liver and Plasma

For metabolomics analyses of the two mouse models, we extracted polar metabolomes of the mouse liver and plasma samples before and after diabetes onset. Metabolome extracts were analyzed in a nontargeted fashion by the FIA TOF MS in negative ionization mode. Upon the sample spectra acquisition, metabolites were identified by matching ions of the unique mass-to-charge (m/z) ratios to those in the KEGG database (48). Since this method cannot distinguish isomers, e.g., isoleucine and leucine, multiple annotations are possible; these annotations are fully disclosed in Supplementary Table 5. Upon processing and annotation, 756 and 755 metabolites were detected and putatively annotated in liver and plasma of β-Phb2−/− mice. The PCA of the liver metabolic profile of the β-Phb2−/− mouse model revealed partial separation between β-Phb2−/− mice and their controls at the age of 4 weeks. This separation became larger with the disease progression (Fig. 4A). Regarding the db/db mouse model, 653 and 595 metabolites ions were detected and putatively annotated in the liver and plasma, respectively. PCA revealed that the liver metabolic profile of db/db mice separated completely from that of the controls at the age of 4 weeks, converged at 6 weeks, and later separated again upon diabetes onset at 8 weeks (Fig. 4B).

Next, we analyzed metabolome changes between β-Phb2−/− and control mice. For each week of development, we performed a univariate analysis to identify significant markers (|log2(FC)| > 1 and q value < 0.01) (Supplementary Table 6). We found 48 and 82 metabolites to be affected in liver and plasma, respectively. A similar differential analysis was done for db/db mice (Supplementary Table 6); 35 and 10 metabolites were filtered in liver and plasma, respectively.

We analyzed the metabolome with a special focus on amino acids and carbohydrate metabolism (Fig. 4C). The most consistent diabetes-related metabolites from human T2D studies are the BCAAs, i.e., valine, leucine, and isoleucine (1214), although their cause-effect relationship with diabetes per se remains unclear. BCAAs have been reported to increase in plasma years before diabetes is diagnosed (13,14); obesity-related insulin resistance probably contributes to this and thus does not reflect β-cell failure. In the plasma of β-Phb2−/− mice, we found a modest increase in BCAAs at the early prediabetic stage (4 weeks) and a more pronounced increase upon the onset of severe diabetes (10 weeks), which might be due to sarcopenia associated with a cachectic state (Fig. 1B and C). In the plasma of fatty db/db mice (Fig. 1E and F), BCAAs markedly increased at the early prediabetic stage (4 weeks), consistent with their association with insulin resistance (49). Apart from BCAAs, two aromatic amino acids, tyrosine and phenylalanine, have been positively associated with diabetes (13,19). Similar to the pattern we observed with BCAAs, the increment of plasma aromatic amino acids was higher in db/db mice than in β-Phb2−/− mice during the prediabetic stage (4 weeks).

Regarding carbohydrate metabolites analyzed by FIA TOF MS (Fig. 4C), hexose sugars (e.g., glucose, mannose, fructose), denoted as glucose (m/z ∼179.056), increased in the plasma of mice from both models as diabetes progressed. Interestingly, in the liver of db/db mice, the increment of hexose sugars paralleled diabetes development, which could be explained by both higher food intake and increased endogenous glucose production caused by insulin resistance. On the contrary, hexose sugars remained unchanged in the liver of β-Phb2−/− mice, except for a slight increase at diabetes onset (6 weeks). Another striking difference between these two mouse models was the alteration in hepatic tricarboxylic acid (TCA) cycle at the prediabetic stage (4 weeks). In the liver of β-Phb2−/− mice, the TCA cycle intermediates fumarate, succinate, and malate slightly increased at the age of 4 weeks. This observation was further substantiated by targeted metabolomics using liquid chromatography–MS (Supplementary Fig. 1). In liver of db/db mice, fumarate, succinate, malate, and citrate dramatically decreased (Fig. 4C), suggesting important hepatic cataplerosis not compensated for by the replenishment of the TCA cycle, i.e., anaplerosis. Malate and citrate exit the TCA cycle to fuel gluconeogenesis and lipogenesis, respectively, potentially contributing to the early-onset insulin resistance in db/db mice. In parallel, the anaplerotic metabolites glutamate, glutamine, and aspartate were reduced in the liver of db/db mice at the age of 4 weeks, indicating lower hepatic anaplerosis.

In summary, alterations of BCAAs, aromatic amino acids, and TCA cycle–related metabolites in prediabetes were much more pronounced in obese db/db mice than in lean β-Phb2−/− mice, pointing to an association of these specific metabolites with insulin resistance rather than β-cell defect per se.

Identification of Metabolites Shared Between β-Phb2−/− and db/db Metabolome Reprogramming

To identify putative metabolites predictive of diabetes development in both db/db and β-Phb2−/− mice, we performed machine learning analysis with the random forests classification algorithm. A randomly sampled training set of data was used for the analysis, and the left out set of data was used for internal cross-validation. The analysis pipeline was conducted as follows: 1) In the first (preprocessing) step, the features displaying the highest correlation with the diabetic state were selected to reduce the dimensionality of the data. 2) In the second (classification) step, the selected metabolites from the β-Phb2−/− cohort were ranked according to their contribution to the classification of the individuals into diabetic and nondiabetic conditions (Table 1). Random forest analysis clearly ranked m/z ∼163.061 as the most important metabolite feature to classify diabetic and nondiabetic animals in both liver and plasma of β-Phb2−/− mice (Supplementary Fig. 2). This m/z corresponds to C6H11O5, i.e., deprotonated deoxyhexose, and might relate to one or multiple isomers such as rhamnose, 1,5-anhydroglucitol, or fucose. Consistent with the machine learning data, the levels of deoxyhexose sugars were significantly lower in the liver and plasma of β-Phb2−/− mice at the euglycemic prediabetic stage (4 weeks) than in their controls (Fig. 5A). Furthermore, the difference of deoxyhexose sugars between β-Phb2−/− and control mice became gradually more pronounced with the development of diabetes. Since the primary tissue alteration upstream of diabetes development in β-Phb2−/− mice is the progressive decline in functional β-cells, deoxyhexose sugars emerged as a direct marker of β-cell mass.

These observations were further substantiated in the db/db mice (Fig. 5B). In their liver, the levels of deoxyhexose sugars were significantly reduced at the age of 6 weeks when mice were still euglycemic. In agreement with the β-Phb2−/− model, the difference in amounts of hepatic deoxyhexose sugars between db/db mice and their controls became larger at the diabetic stage 2 weeks later (8 weeks). In the plasma of db/db mice, marginal differences were observed in the prediabetic stage, whereas significant changes appeared with diabetes (8 weeks).

We then performed targeted metabolomics by GC-MS to disambiguate the molecular identity of the m/z ∼163.061 deoxyhexose sugar that contributes most to the differentiation at the prediabetic stage. Metabolite profiling by GC-MS identifies isomers with identical molecular mass but different structures by differential chemical derivatization and gas chromatographic separation. Analyses performed on plasma and liver samples by using conventional GC/electron ionization TOF MS and enhanced GC/atmospheric pressure chemical ionization quadrupole TOF MS (39,41) led to the identification of two analytes: 1,5-anhydroglucitol (Fig. 6A) and fucose (Fig. 7A). This corresponds to m/z ∼163.061 ion that was detected by FIA TOF MS. Consistent with nontargeted metabolomics results from TOF MS, the level of 1,5-anhydroglucitol was significantly lower in the liver and plasma of β-Phb2−/− mice at the prediabetic stage (5 weeks) than in their controls (Fig. 6B); this difference became more pronounced as diabetes progressed. In the liver of db/db mice (Fig. 6C), 1,5-anhydroglucitol levels were also significantly reduced at the early diabetic stage (8 weeks). However, at the prediabetic stage, the difference between db/db mice and their controls did not reach significance (P = 0.093). In light of these results, we took advantage of previously published studies reporting 1,5-anhydroglucitol plasma levels in alternative rodent models of diabetes secondary to β-cell loss. NOD mice spontaneously develop diabetes by the age of 15–25 weeks (50). A decrease in plasma 1,5-anhydroglucitol was consistently observed 10–30 days before the hyperglycemia (51) (see Supplementary Fig. 3A). Regarding toxin-induced diabetes, injection of streptozotocin exerts a rapid cytotoxic action on β-cells, inducing diabetes within hours, which was preceded by a decline in 1,5-anhydroglucitol plasma levels (52) (see Supplementary Fig. 3B).

For both db/db and β-Phb2−/− mouse models, plasma levels of the m/z ∼163.061 deoxyhexose fucose were unchanged, while there were some differences in their livers (Fig. 7B and C). In β-Phb2−/− mice at the prediabetic stage (4 weeks), fucose levels were slightly higher in the liver compared with controls—a difference that vanished with time (Fig. 7B). For db/db mice, fucose levels were lower in the liver at the prediabetic stage (6 weeks) and remained lower than those in controls when mice became hyperglycemic (Fig. 7C). Thus, as opposed to levels of its isomer 1,5-anhydroglucitol, fucose levels did not correlate with β-cell failure.

In conclusion, both β-Phb2−/− and db/db mouse models exhibited early changes specifically in the deoxyhexose 1,5-anhydroglucitol, in association with the development of diabetes triggered by a decline of functional β-cell mass.

A sensitive and robust biomarker of functional β-cell mass is an unmet medical need. To date, noninvasive imaging of total β-cell mass is not yet available at the clinical level (53). Using metabolomics in two mouse models with complementary and well-characterized spectra of β-cell defects, we identified the deoxyhexose 1,5-anhydroglucitol as a biomarker that closely associated with functional β-cell mass preceding the appearance of hyperglycemia. Specifically, the liver and plasma levels of 1,5-anhydroglucitol lowered along with functional β-cell mass, through an unidentified mechanism, preceding diabetes onset and progressing further as diabetes manifested to a severe stage. Consistently, inverse association of deoxyhexose sugars with T2D about 6 years before T2D diagnosis has been reported in an untargeted study of a German cohort (54). In this cohort, the deoxyhexose levels were poorly correlated with other risk factors, such as age and BMI (54). In line with these results, a Korean study showed that low serum levels of 1,5-anhydroglucitol in euglycemic prediabetic subjects were associated with reduced insulinogenic index (a metric of insulin secretion capacity) and not with higher HOMA-IR (55). Regarding type 1 diabetes, a recent GC-MS analysis identified circulating metabolites in children who later progress to autoimmune diabetes, showing lower plasma levels of the sugar class 1,5-anhydrohexitol (e.g., 1,5-anhydroglucitol) at 6 months of age than are found in control subjects (56).

The deoxyhexose 1,5-anhydroglucitol has previously been reported as an index of short-term glycemic control in patients with T2D (57). Similar changes in 1,5-anhydroglucitol have been measured in ob/ob and db/db mice (58). In human cohorts, 1,5-anhydroglucitol has been reported to be 38% lower in patients with diabetes than in control subjects (59), and the deoxyhexose correlated with impaired fasting glycemia and T2D (14,15). 1,5-Anhydroglucitol is a naturally occurring polyol found in nearly all foods and hardly metabolized in humans (45). It can also be synthesized from glycogen in the liver (60), although its de novo synthesis accounts for a small percentage of its whole-body storage (61). The primary route for 1,5-anhydroglucitol disposal is urinary excretion, while hyperglycemia has been shown to promote such a mechanism, resulting in the lowering of plasma levels of 1,5-anhydroglucitol (45,62). Indeed, 1,5-anhydroglucitol can be reabsorbed by the sodium–glucose cotransporter 4 in renal tubules (63), but its reabsorption is inhibited when blood glucose levels reach 10 mmol/L. Accordingly, it has been suggested that 1,5-anhydroglucitol is sensitive in reflecting postprandial glycemic excursions (64). The impaired glucose clearance in β-Phb2−/− mice secondary to β-cell failure (26) may contribute to reduced renal reabsorption of 1,5-anhydroglucitol and lowering of its plasma level. In parallel, the decrease in hepatic 1,5-anhydroglucitol concentrations observed in β-Phb2−/− mice could be the consequence of lower glycogen-derived biosynthesis, reducing its efflux normally occurring across the cell membrane (60,65). These mechanisms may contribute to the lowering in plasma 1,5-anhydroglucitol levels, reflecting the progressive decline of functional β-cell mass in β-Phb2−/− mice. Of note, we did not analyze the metabolome of the β-cells, since pancreata were fixed for the assessment of the β-cell mass by immunohistochemistry in parallel to liver and plasma metabolomics. Therefore, the changes in liver and plasma 1,5-anhydroglucitol might be induced indirectly by the β-cell decline, possibly by early modifications of hepatic metabolism in response to a prediabetic stage. Another constraint in our study was the limited amount of samples, in particular plasma volumes. We therefore focused on a single analytical metabolomics approach and negative ionization. In the future, the breadth of the analysis could be expanded by, for example, including positive ionization or complementary methods for analyzing lipid extracts.

The vast heterogeneity of human genetics cannot be recapitulated in one mouse model, although investigating a highly heterogeneous background may introduce multiple confounding factors precluding the identification of a metabolite of significance. To reduce the putative impact of the genetic background and to widen the mechanisms promoting diabetes, we extended our analyses to db/db mice. They are characterized by β-cell mass expansion (28,29) as a sign of compensation for insulin resistance. However, such expansion of insulin-positive cells does not prevent the development of hyperglycemia, observed at the age of 8 weeks, as a consequence of β-cell dysfunction. Glucose-stimulated calcium responses and insulin exocytosis are reduced in prediabetic db/db mice (30). The present study reveals similar changes in deoxyhexose levels in prediabetic db/db and β-Phb2−/− mice, which are less pronounced in the former. BCAAs and aromatic amino acids are recognized predictors of insulin resistance (12,66). Accordingly, our data revealed higher increments of these amino acids in the plasma of overweight db/db mice than in lean β-Phb2−/− mice. These changes indicate that insulin resistance in db/db mice has a great impact on the metabolome in prediabetes, not directly related to β-cell mass.

In conclusion, the common denominator for prediabetic db/db and β-Phb2−/− mice was restricted to 1,5-anhydroglucitol, a sugar that can be easily measured by enzyme-based colorimetric assay. This deoxyhexose reflects progressive decline of functional β-cell mass at the asymptomatic prediabetic stage. Although this points to a valuable biomarker of early asymptomatic β-cell loss, validation of our findings in human cohorts would be the ultimate goal. If combined with already known clinical risk factors, such as BMI and fasting glucose, such a biomarker could add value in predicting progression to T2D and could enable tailored treatment strategies.

L.L. and P.K. contributed equally to this work.

Acknowledgments. The authors thank Christian Vesin and Florian Visentin for animal handling and sample collection, Clarissa Bartley for performing the insulin secretion experiment, Luisa Carvalho for performing histology, and Gaelle Chaffard for extracting metabolites (all from University of Geneva, Geneva, Switzerland). The authors are grateful to Vincent Deo (Stanford University, Stanford, California) for helping with the machine learning analysis. The authors thank the members of the bioimaging and histology core facilities (University of Geneva, Geneva, Switzerland) for providing technical support.

Funding. This study was financed by grants from Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (the Swiss National Science Foundation) (Sinergia #CRSII3_147637 to N.Z. and P.M. and #P1GEP3_161866 to A.A.).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. L.L., P.K., A.A., N.Z., and P.M. analyzed the data. L.L., P.K., and P.M. wrote the manuscript. L.L., J.M.-L., and S.S. performed experiments on mice and related parameters. L.L., N.Z., and P.M. conceived and designed the experiments. P.K. performed the metabolomics. A.E., A.A., J.K., and N.Z. made contributions to the manuscript. A.E. and J.K. performed and analyzed the targeted metabolomics based on gas chromatography–mass spectrometry. P.M. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

1.
DiMeglio
LA
,
Evans-Molina
C
,
Oram
RA
.
Type 1 diabetes
.
Lancet
2018
;
391
:
2449
2462
2.
Chatterjee
S
,
Khunti
K
,
Davies
MJ
.
Type 2 diabetes
.
Lancet
2017
;
389
:
2239
2251
3.
Halban
PA
,
Polonsky
KS
,
Bowden
DW
, et al
.
β-Cell failure in type 2 diabetes: postulated mechanisms and prospects for prevention and treatment
.
Diabetes Care
2014
;
37
:
1751
1758
4.
Ashcroft
FM
,
Rorsman
P
.
Diabetes mellitus and the β cell: the last ten years
.
Cell
2012
;
148
:
1160
1171
5.
Matveyenko
AV
,
Butler
PC
.
Relationship between beta-cell mass and diabetes onset
.
Diabetes Obes Metab
2008
;
10
(
Suppl. 4
):
23
31
6.
Butler
AE
,
Janson
J
,
Bonner-Weir
S
,
Ritzel
R
,
Rizza
RA
,
Butler
PC
.
Beta-cell deficit and increased beta-cell apoptosis in humans with type 2 diabetes
.
Diabetes
2003
;
52
:
102
110
7.
Del Guerra
S
,
Lupi
R
,
Marselli
L
, et al
.
Functional and molecular defects of pancreatic islets in human type 2 diabetes
.
Diabetes
2005
;
54
:
727
735
8.
Mari
A
,
Ferrannini
E
.
Beta-cell function assessment from modelling of oral tests: an effective approach
.
Diabetes Obes Metab
2008
;
10
(
Suppl. 4
):
77
87
9.
Remedi
MS
,
Emfinger
C
.
Pancreatic β-cell identity in diabetes
.
Diabetes Obes Metab
2016
;
18
(
Suppl. 1
):
110
116
10.
Adamski
J
.
Key elements of metabolomics in the study of biomarkers of diabetes
.
Diabetologia
2016
;
59
:
2497
2502
11.
Fuhrer
T
,
Zamboni
N
.
High-throughput discovery metabolomics
.
Curr Opin Biotechnol
2015
;
31
:
73
78
12.
Newgard
CB
,
An
J
,
Bain
JR
, et al
.
A branched-chain amino acid-related metabolic signature that differentiates obese and lean humans and contributes to insulin resistance
.
Cell Metab
2009
;
9
:
311
326
13.
Wang
TJ
,
Larson
MG
,
Vasan
RS
, et al
.
Metabolite profiles and the risk of developing diabetes
.
Nat Med
2011
;
17
:
448
453
14.
Yu
D
,
Moore
SC
,
Matthews
CE
, et al
.
Plasma metabolomic profiles in association with type 2 diabetes risk and prevalence in Chinese adults
.
Metabolomics
2016
;
12
.
pii. 3
15.
Menni
C
,
Fauman
E
,
Erte
I
, et al
.
Biomarkers for type 2 diabetes and impaired fasting glucose using a nontargeted metabolomics approach
.
Diabetes
2013
;
62
:
4270
4276
16.
Peddinti
G
,
Cobb
J
,
Yengo
L
, et al
.
Early metabolic markers identify potential targets for the prevention of type 2 diabetes
.
Diabetologia
2017
;
60
:
1740
1750
17.
Gall
WE
,
Beebe
K
,
Lawton
KA
, et al.;
RISC Study Group
.
Alpha-hydroxybutyrate is an early biomarker of insulin resistance and glucose intolerance in a nondiabetic population
.
PLoS One
2010
;
5
:
e10883
18.
Ferrannini
E
,
Natali
A
,
Camastra
S
, et al
.
Early metabolic markers of the development of dysglycemia and type 2 diabetes and their physiological significance
.
Diabetes
2013
;
62
:
1730
1737
19.
Floegel
A
,
Stefan
N
,
Yu
Z
, et al
.
Identification of serum metabolites associated with risk of type 2 diabetes using a targeted metabolomic approach
.
Diabetes
2013
;
62
:
639
648
20.
Wang-Sattler
R
,
Yu
Z
,
Herder
C
, et al
.
Novel biomarkers for pre-diabetes identified by metabolomics
.
Mol Syst Biol
2012
;
8
:
615
21.
Guasch-Ferré
M
,
Hruby
A
,
Toledo
E
, et al
.
Metabolomics in prediabetes and diabetes: a systematic review and meta-analysis
.
Diabetes Care
2016
;
39
:
833
846
22.
Klein
MS
,
Shearer
J
.
Metabolomics and type 2 diabetes: translating basic research into clinical application
.
J Diabetes Res
2016
;
2016
:
3898502
23.
Stumvoll
M
,
Mitrakou
A
,
Pimenta
W
, et al
.
Use of the oral glucose tolerance test to assess insulin release and insulin sensitivity
.
Diabetes Care
2000
;
23
:
295
301
24.
Basu
R
,
Schwenk
WF
,
Rizza
RA
.
Both fasting glucose production and disappearance are abnormal in people with “mild” and “severe” type 2 diabetes
.
Am J Physiol Endocrinol Metab
2004
;
287
:
E55
E62
25.
Krogvold
L
,
Edwin
B
,
Buanes
T
, et al
.
Pancreatic biopsy by minimal tail resection in live adult patients at the onset of type 1 diabetes: experiences from the DiViD study
.
Diabetologia
2014
;
57
:
841
843
26.
Supale
S
,
Thorel
F
,
Merkwirth
C
, et al
.
Loss of prohibitin induces mitochondrial damages altering β-cell function and survival and is responsible for gradual diabetes development
.
Diabetes
2013
;
62
:
3488
3499
27.
Lee
GH
,
Proenca
R
,
Montez
JM
, et al
.
Abnormal splicing of the leptin receptor in diabetic mice
.
Nature
1996
;
379
:
632
635
28.
Kodama
H
,
Fujita
M
,
Yamaguchi
I
.
Development of hyperglycaemia and insulin resistance in conscious genetically diabetic (C57BL/KsJ-db/db) mice
.
Diabetologia
1994
;
37
:
739
744
29.
Dalbøge
LS
,
Almholt
DL
,
Neerup
TS
, et al
.
Characterisation of age-dependent beta cell dynamics in the male db/db mice
.
PLoS One
2013
;
8
:
e82813
30.
Do
OH
,
Gunton
JE
,
Gaisano
HY
,
Thorn
P
.
Changes in beta cell function occur in prediabetes and early disease in the Lepr (db) mouse model of diabetes
.
Diabetologia
2016
;
59
:
1222
1230
31.
Meier
JJ
,
Breuer
TG
,
Bonadonna
RC
, et al
.
Pancreatic diabetes manifests when beta cell area declines by approximately 65% in humans
.
Diabetologia
2012
;
55
:
1346
1354
32.
Fuhrer
T
,
Heer
D
,
Begemann
B
,
Zamboni
N
.
High-throughput, accurate mass metabolome profiling of cellular extracts by flow injection-time-of-flight mass spectrometry
.
Anal Chem
2011
;
83
:
7074
7080
33.
Karaca
M
,
Martin-Levilain
J
,
Grimaldi
M
, et al
.
Liver glutamate dehydrogenase controls whole-body energy partitioning through amino acid-derived gluconeogenesis and ammonia homeostasis
.
Diabetes
2018
;
67
:
1949
1961
34.
Izumida
Y
,
Yahagi
N
,
Takeuchi
Y
, et al
.
Glycogen shortage during fasting triggers liver-brain-adipose neurocircuitry to facilitate fat utilization
.
Nat Commun
2013
;
4
:
2316
35.
Matthews
DR
,
Hosker
JP
,
Rudenski
AS
,
Naylor
BA
,
Treacher
DF
,
Turner
RC
.
Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man
.
Diabetologia
1985
;
28
:
412
419
36.
Carobbio
S
,
Ishihara
H
,
Fernandez-Pascual
S
,
Bartley
C
,
Martin-Del-Rio
R
,
Maechler
P
.
Insulin secretion profiles are modified by overexpression of glutamate dehydrogenase in pancreatic islets
.
Diabetologia
2004
;
47
:
266
276
37.
Storey
JD
.
A direct approach to false discovery rates
.
J R Stat Soc B
2002
;
64
:
479
498
38.
Buescher
JM
,
Moco
S
,
Sauer
U
,
Zamboni
N
.
Ultrahigh performance liquid chromatography-tandem mass spectrometry method for fast and robust quantification of anionic and aromatic metabolites
.
Anal Chem
2010
;
82
:
4403
4412
39.
Dethloff
F
,
Erban
A
,
Orf
I
, et al
.
Profiling methods to identify cold-regulated primary metabolites using gas chromatography coupled to mass spectrometry
.
Methods Mol Biol
2014
;
1166
:
171
197
40.
Erban
A
,
Schauer
N
,
Fernie
AR
,
Kopka
J
.
Nonsupervised construction and application of mass spectral and retention time index libraries from time-of-flight gas chromatography-mass spectrometry metabolite profiles
.
Methods Mol Biol
2007
;
358
:
19
38
41.
Kopka
J
,
Schmidt
S
,
Dethloff
F
, et al
.
Systems analysis of ethanol production in the genetically engineered cyanobacterium Synechococcus sp. PCC 7002
.
Biotechnol Biofuels
2017
;
10
:
56
42.
Strehmel
N
,
Hummel
J
,
Erban
A
,
Strassburg
K
,
Kopka
J
.
Retention index thresholds for compound matching in GC-MS metabolite profiling
.
J Chromatogr B Analyt Technol Biomed Life Sci
2008
;
871
:
182
190
43.
Luedemann
A
,
Strassburg
K
,
Erban
A
,
Kopka
J
.
TagFinder for the quantitative analysis of gas chromatography--mass spectrometry (GC-MS)-based metabolite profiling experiments
.
Bioinformatics
2008
;
24
:
732
737
44.
Hummel
J
,
Strehmel
N
,
Selbig
J
,
Walther
D
,
Kopka
J
.
Decision tree supported substructure prediction of metabolites from GC-MS profiles
.
Metabolomics
2010
;
6
:
322
333
45.
Yamanouchi
T
,
Tachibana
Y
,
Akanuma
H
, et al
.
Origin and disposal of 1,5-anhydroglucitol, a major polyol in the human body
.
Am J Physiol
1992
;
263
:
E268
E273
46.
Hastie
T
,
Tibshirani
R
,
Friedman
J
.
The Elements of Statistical Learning
.
New York, NY
,
Springer New York Inc.
,
2009
47.
Pedregosa
F
,
Varoquaux
G
,
Gramfort
A
, et al
.
Scikit-learn: machine learning in Python
.
J Mach Learn Res
2011
;
12
:
2825
2830
48.
Kanehisa
M
.
Enzyme annotation and metabolic reconstruction using KEGG
.
Methods Mol Biol
2017
;
1611
:
135
145
49.
Mahendran
Y
,
Jonsson
A
,
Have
CT
, et al
.
Genetic evidence of a causal effect of insulin resistance on branched-chain amino acid levels
.
Diabetologia
2017
;
60
:
873
878
50.
Mathews
CE
,
Xue
S
,
Posgai
A
, et al
.
Acute versus progressive onset of diabetes in NOD mice: potential implications for therapeutic interventions in type 1 diabetes
.
Diabetes
2015
;
64
:
3885
3890
51.
Kametani
S
,
Hashimoto
Y
,
Yamanouchi
T
,
Akanuma
Y
,
Akanuma
H
.
Reduced renal reabsorption of 1,5-anhydro-D-glucitol in diabetic rats and mice
.
J Biochem
1987
;
102
:
1599
1607
52.
Yamanouchi
T
,
Akanuma
H
,
Takaku
F
,
Akanuma
Y
.
Marked depletion of plasma 1,5-anhydroglucitol, a major polyol, in streptozocin-induced diabetes in rats and the effect of insulin treatment
.
Diabetes
1986
;
35
:
204
209
53.
Gotthardt
M
,
Eizirik
DL
,
Cnop
M
,
Brom
M
.
Beta cell imaging - a key tool in optimized diabetes prevention and treatment
.
Trends Endocrinol Metab
2014
;
25
:
375
377
54.
Drogan
D
,
Dunn
WB
,
Lin
W
, et al
.
Untargeted metabolic profiling identifies altered serum metabolites of type 2 diabetes mellitus in a prospective, nested case control study
.
Clin Chem
2015
;
61
:
487
497
55.
Won
JC
,
Park
CY
,
Park
HS
, et al
.
1,5-Anhydroglucitol reflects postprandial hyperglycemia and a decreased insulinogenic index, even in subjects with prediabetes and well-controlled type 2 diabetes
.
Diabetes Res Clin Pract
2009
;
84
:
51
57
56.
Lamichhane
S
,
Kemppainen
E
,
Trošt
K
, et al
.
Circulating metabolites in progression to islet autoimmunity and type 1 diabetes
.
Diabetologia
.
23 August
2019
[Epub ahead of print]
. DOI: 10.1007/s00125-019-04980-0
57.
McGill
JB
,
Cole
TG
,
Nowatzke
W
, et al.;
U.S. trial of the GlycoMark assay
.
Circulating 1,5-anhydroglucitol levels in adult patients with diabetes reflect longitudinal changes of glycemia: a U.S. trial of the GlycoMark assay
.
Diabetes Care
2004
;
27
:
1859
1865
58.
Giesbertz
P
,
Padberg
I
,
Rein
D
, et al
.
Metabolite profiling in plasma and tissues of ob/ob and db/db mice identifies novel markers of obesity and type 2 diabetes
.
Diabetologia
2015
;
58
:
2133
2143
59.
Suhre
K
,
Meisinger
C
,
Döring
A
, et al
.
Metabolic footprint of diabetes: a multiplatform metabolomics study in an epidemiological setting
.
PLoS One
2010
;
5
:
e13953
60.
Kametani
S
,
Shiga
Y
,
Akanuma
H
.
Hepatic production of 1,5-anhydrofructose and 1,5-anhydroglucitol in rat by the third glycogenolytic pathway
.
Eur J Biochem
1996
;
242
:
832
838
61.
Kim
WJ
,
Park
CY
.
1,5-Anhydroglucitol in diabetes mellitus
.
Endocrine
2013
;
43
:
33
40
62.
Yamanouchi
T
,
Shinohara
T
,
Ogata
N
,
Tachibana
Y
,
Akaoka
I
,
Miyashita
H
.
Common reabsorption system of 1,5-anhydro-D-glucitol, fructose, and mannose in rat renal tubule
.
Biochim Biophys Acta
1996
;
1291
:
89
95
63.
Tazawa
S
,
Yamato
T
,
Fujikura
H
, et al
.
SLC5A9/SGLT4, a new Na+-dependent glucose transporter, is an essential transporter for mannose, 1,5-anhydro-D-glucitol, and fructose
.
Life Sci
2005
;
76
:
1039
1050
64.
Dungan
KM
,
Buse
JB
,
Largay
J
, et al
.
1,5-anhydroglucitol and postprandial hyperglycemia as measured by continuous glucose monitoring system in moderately controlled patients with diabetes
.
Diabetes Care
2006
;
29
:
1214
1219
65.
Suzuki
M
,
Akanuma
H
,
Akanuma
Y
.
Transport of 1,5-anhydro-D-glucitol across plasma membranes in rat hepatoma cells
.
J Biochem
1988
;
104
:
956
959
66.
Tai
ES
,
Tan
ML
,
Stevens
RD
, et al
.
Insulin resistance is associated with a metabolic profile of altered protein metabolism in Chinese and Asian-Indian men
.
Diabetologia
2010
;
53
:
757
767
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at http://www.diabetesjournals.org/content/license.

Supplementary data