Plasma Fucosylated Glycans and C-Reactive Protein as Biomarkers of HNF1A-MODY in Young Adult–Onset Nonautoimmune Diabetes

Although a number of genes are implicated in monogenic diabetes, maturity-onset diabetes of the young (MODY) due to variants in HNF1A (HNF1A-MODY) is the most frequent form in adults (1) and has a significant effect on management when the diagnosis is made. Common clinical criteria for selecting individuals for genetic testing for MODY include diabetes onset younger than 25 years of age, preserved endogenous insulin production, absence of pancreatic autoimmunity, and consecutive generations of diabetes (2,3).

These criteria clearly overlap with features of both type 1 and type 2 diabetes, so many individuals with HNF1A-MODY remain unrecognized, particularly if they do not fit classic MODY criteria (1). Herein, we defined HNF1A-MODY as nonautoimmune young adult–onset diabetes in individuals carrying deleterious HNF1A alleles.

Frequently, physicians do not prioritize the diagnosis of MODY, and also, many countries have limited access to genetic testing. Establishing a correct molecular diagnosis of HNF1A-MODY allows treatment change to sulfonylureas or glinides, which may provide excellent diabetes control for decades (4). The correct diagnosis also facilitates prompt identification of affected family members.

Widening access to MODY genetic testing is assisted by tools such as the MODY probability calculator (5); however, these models rely largely on clinical criteria. Criteria such as absence of β-cell autoantibodies and presence of C-peptide exclude most cases of autoimmune diabetes. Work in this area has focused mainly on the discrimination of MODY presenting in childhood by selecting β-cell antibody–negative children for further investigation (6). A recent study took a similar approach with young adults diagnosed before 30 years (7), although most had a clinical label of type 1 diabetes.

The differentiation from type 2 diabetes, particularly in an older age group where the proportion of MODY is lower, is more challenging. Adding specific HNF1A-MODY biomarkers that rely on extrapancreatic manifestations of HNF1A could assist in the differentiation from nonautoimmune diabetes.

HNF1A encodes a transcription factor that regulates the expression of many genes (8). C-reactive protein (CRP) expression in the liver is regulated by HNF1A (9) and genome-wide association studies (GWAS) showed that the plasma CRP level was associated with genetic variation near HNF1A (10,11). We subsequently reported that high-sensitivity CRP (hs-CRP) levels were lower in subjects with typical HNF1A-MODY than in other forms of diabetes, with the best discrimination from young adult–onset type 2 diabetes (12).

Similarly, a GWAS of the plasma N-glycome identified HNF1A as a key regulator of plasma protein fucosylation (13). N-glycosylation is a frequent posttranslational modification, characterized by enzymatic attachment of complex sugar moieties (glycans) to the protein. It is essential for proper protein function (14) and has an important role in many (patho)physiological processes (15,16). We also reported that disease-causing HNF1A alleles are associated with marked alterations of plasma N-glycans bearing antennary fucose (17).

Thus, plasma N-glycans and hs-CRP are both promising candidates for HNF1A-MODY diagnostic markers. Our previous studies focused on individuals with an established clinical diagnosis based on clinical, biochemical, and molecular investigations. These studies were likely to be subject to spectrum bias, leading to an overestimation of the discriminative properties.

Our aim in this study was to assess the value of N-glycans and hs-CRP as HNF1A-MODY biomarkers in a relatively unselected population with a young adult–onset nonautoimmune diabetes and evaluate their translational potential. As part of this process, evaluating whether the identified rare HNF1A alleles were likely to be the main cause of diabetes in the individuals recruited was necessary.

Subjects were recruited in U.K. and Croatia. U.K. participants (n = 523) were recruited via the Young Diabetes in Oxford (YDX) study, which included seven hospital diabetes centers and multiple primary care centers from the Thames Valley. Croatian subjects (n = 466) were recruited through the Croatian National Diabetes Registry (CroDiab) and sampled at Vuk Vrhovac University Clinic in Zagreb.

Inclusion criteria were current age ≥18 years, diabetes diagnosis at <45 years, preserved endogenous insulin production (fasting C-peptide ≥0.2 nmol/L), and negative GAD antibodies (GADAs). From 989 subjects included in the study, 84 had diabetes onset before 25 years of age. Four U.K. subjects had a previously known diagnosis of HNF1A-MODY as a result of previous investigation in the YDX study. All participants signed an informed consent. Table 1 summarizes the clinical characteristics of the recruited individuals and their treatment at the time of recruitment.

DNA was extracted, amplified, and sequenced using the Sanger method (18). Mutation Surveyor version 5.0.1 (SoftGenetics, Cambridge, U.K.) was used for detection of variants compared with the reference sequence (NM_000545.5). A systematic assessment of rare HNF1A alleles (minor allele frequency [MAF] <1%) was performed and aligned to the American College of Medical Genetics (ACMG) classification (19). This included clinical features, cosegregation of the allele with diabetes in the family, and in silico analysis of missense variants using sorting intolerant from tolerant (SIFT) and Polymorphism Phenotyping version 2 (PolyPhen2), and Protein Variation Effect Analyzer (PROVEAN). Potential effect on splicing was examined using Human Splicing Finder (HSF). The presence of rare HNF1A alleles in the publicly available database of 123,136 exomes and 15,496 whole-genomes in Genome Aggregation Database (GnomAD; Broad Institute, Cambridge, MA) (20) was recorded. The results of laboratory assessment of function, available in the literature, were reviewed, and functional studies of five previously uncharacterized HNF1A alleles were performed.

cDNA of human HNF1A (NM_000545.5) was inserted into the pcDNA 3.1/His C plasmid (gift from KG Jebsen Center for Diabetes Research, University of Bergen) and used in site-directed mutagenesis as a template. HeLa cells were cultured for all functional assessments and transfected with mutagenized plasmids. Each HNF1A variant underwent an assessment of transcription activity using a dual-luciferase reporter system, an analysis of protein expression using Western blotting, and an assessment of DNA binding using electrophoretic mobility shift assay (EMSA). Each experiment included empty plasmid, wild-type (WT) HNF1A, two to three positive controls (p.P112L, p.T260M, and p.R203H), one synonymous variant (p.H179H), and one variant associated with an increased risk of type 2 diabetes (p.E508K). The choice of control variants was based on an established evidence for causing MODY, such as cosegregation of the allele with the young adult–onset diabetes in multiple families and supporting functional data. Each experiment was repeated twice on further passages of HeLa cells to obtain three biological replicates.

N-Glycan release, labeling, and cleanup was performed as described previously (21). Fluorescently labeled glycans were separated by hydrophilic interaction liquid chromatography on an Acquity UPLC instrument (Waters, Milford, MA), as previously described (21). All chromatograms were separated into 42 chromatographic peaks, which enabled reliable quantification. The amount of glycans in each peak was expressed as the percentage of the total integrated area. The corresponding glycan structures were assigned according to Saldova et al. (22).

CRP was measured in most of the U.K. samples (n = 495) by using a wide-range latex-enhanced immunoturbidimetric high-sensitivity assay on an ADVIA 2400 analyzer (Siemens Healthcare Diagnostics, Erlangen, Germany) with the limit of quantification of 0.01 mg/L. CRP measurements for all Croatian samples and the remaining U.K. samples (n = 494) were done with the Abbott hs-CRP method and the lowest quantifiable level of 0.1 mg/L. Methods were reproducible, with the coefficient of variation for both below 10.5% across the concentration range tested. Comparison of the clinical samples (n = 51) measured by both methods showed agreement to be Abbott method = 0.26 + 0.99 (Siemens method), by Passing and Bablock regression.

Measurements of GADAs were performed in laboratories participating in the Diabetes Antibody Standardization Program (DASP) (23). In U.K., GADAs were measured by radioimmunoassay using 35S-labeled GAD65. The cutoff for the positive result was 13 World Health Organization units/mL initially using a local assay (samples measured n = 218; DASP 2010 sensitivity 88% at 93% specificity) and changed to 33 digestive and kidney units/mL later in the study (standard assay, DASP 2010 sensitivity 80%, specificity 97%). In Croatia, GADAs were measured by ELISA immunoassay. The cutoff for the positive result was 5 World Health Organization units/mL (DASP 2010 sensitivity 88% at 94% specificity).

Subject characteristics and results of the functional work were analyzed using SPSS 23 software. Continuous data are presented as medians with the interquartile range (IQR), and the Kruskal-Wallis test was applied to compare groups. A P value of <0.05 was considered statistically significant. Differences of frequencies for categorical variables were tested using the χ² test. The results of the functional studies are presented as a percentage of the WT HNF1A. Differences between the studied variants and WT were analyzed using ANOVA with correction for multiple tests.

The Exeter MODY probability calculator (5) was used to compare performance of the biomarkers assessed in this study.

Glycan and hs-CRP data were analyzed and visualized using R 3.0.1 software. N-glycans and hs-CRP both had nonparametric distributions. Association analyses between disease status and glycan traits were performed using a general linear model, with age and sex included as additional covariates. The false discovery rate was accounted for using the Benjamini-Hochberg procedure. For prediction of disease status, logistic regression and regularized logistic regression models were both applied. The logistic model was applied in bivariate regression classification analyses (one glycan trait used per model). In multiple regression classification analyses (multiple glycan traits used as predictors in the model), regularized logistic models were applied. To evaluate the performance of the regularized logistic model, the 10-cross validation procedure was used. Predictions from each validation procedure were merged into one validation set on which model performance was evaluated, based on the receiver operating characteristic (ROC) curve criteria.

HNF1A sequencing of 989 study participants resulted in identifying 25 rare (MAF <1%) nonsynonymous HNF1A variants in 29 probands, including 7 protein-truncating variants (PTVs) and 18 missense variants. The identified variants are listed in Table 2. Additional features of all probands are listed in Supplementary Table 1.

The likely phenotypical effect of the HNF1A allele (damaging, variants of unknown significance [VUS], or benign) was assigned taking into consideration previous reports of the allele causing the MODY phenotype, cosegregation of the allele with diabetes, prediction of bioinformatics, absence of the allele in GnomAD, and results of functional studies.

Published data showed that 14 alleles were previously reported as causing the MODY phenotype (24–32), 8 were reported as VUS or benign (24,33), and 3 were novel (p.S3C, p.G151S, p.K222N).

PTVs were all located in exons 1–6, affecting all isoforms of the protein and likely to undergo nonsense-mediated decay leading to haploinsufficiency. All PTVs were previously reported to cause the MODY phenotype with evidence of cosegregation of the allele with diabetes (Table 2). We therefore considered all PTVs as deleterious.

Of the 18 missense variants, 7 were predicted to be damaging protein function by at least two of the three bioinformatic tools used, and 11 were predicted to be benign. HSF predicted that the promoter variant c.-4A>G does not affect splicing. HSF also predicted that the novel coding missense variant c.8C>G, p.S3C, is likely to affect splicing by gaining a new donor site. Bioinformatics also predicted it as damaging.

Of the 25 alleles, 13 were present in individuals from GnomAD with MAF of 0.0008–0.08%, and 12 were not reported in the GnomAD database.

We sequenced HNF1A in 22 available family members from 8 families. Five rare HNF1A alleles were present in eight individuals. Cosegregation of the HNF1A allele and diabetes was reported in this study and/or published literature for 14 of 25 alleles (Table 2).

Five HNF1A protein variants underwent functional assessment: two were novel (p.G151S and p.K222N) and three were not previously studied (p.G288W, p.P291T, and p.H349Q). During this project, an assessment of p.H349Q was published (34), and findings were consistent with ours.

The protein expression of p.G151S, p.G288W, p.P291T, and p.H349Q was similar to the WT HNF1A (107–128% of WT, nonsignificant P values) and increased for p.K222N (P = 0.004) (Fig. 1B).

The normalized transcription activity (TA) was significantly reduced by p.P291T (52% of WT, P = 0.001) and p.G288W (73% of WT, P = 0.04), but the difference in TA was not significant for p.G151S and p.K222N (82.8% and 76.1% of the WT, respectively). The transactivation of p.H349Q was similar to the WT (111% WT) (Fig. 1A).

Finally, we performed EMSA to assess DNA binding of the variants, of which p.G151S and p.K222N are located in the DNA-binding domain. The variant p.G151S had significantly reduced DNA binding (19% of the WT, P = 2 × 10⁻⁶) (Fig. 1C), which remained the same after normalization to the protein amount (Fig. 1D). Normalized DNA binding of the remaining four variants was not different from WT (71–101% of WT).

In summary, the functional assessment of five previously uncharacterized HNF1A protein variants provided support for the novel variant p.G151S to be considered as functionally deleterious. The borderline reduction in TA of variant p.P291T and p.G288 W makes the functional results inconclusive.

Based on the systematic assessment described above and summarized in Table 2, we considered that 12 rare HNF1A alleles (present in 16 probands and 3 relatives) are likely to be damaging HNF1A protein function (ACMG classification 1–2) (19), 9 are likely to be benign (ACMG 4–5), and 4 were labeled as VUS (ACMG 3), because there were features both for and against a damaging effect. This corroborates the observation that the phenotypical effect of the HNF1A alleles represents a spectrum without clear borders and shows the complexity of interpretation of the genetic variation. Although acknowledging this complexity, we simplified the phenotypical spectrum to likely damaging, VUS, and likely benign alleles to enable assessment of the biomarkers.

We excluded 2 individuals with benign alleles due to missing data, leaving 27 probands for N-glycan data analysis. We also excluded 114 subjects (including 2 with benign alleles) with CRP >10 mg/L from the hs-CRP data analysis because hs-CRP may have been elevated as a result of concomitant inflammation.

Firstly, we compared probands with likely damaging HNF1A alleles (n = 16) with individuals without rare HNF1A alleles (n = 960 for glycan and n = 844 for hs-CRP analysis). We found that 8 of 42 glycan traits were significantly altered in subjects with likely damaging HNF1A alleles compared with subjects without HNF1A variants (adjusted P = 1.00 × 10⁻⁵ to 1.46 × 10⁻²) (Supplementary Table 2). Glycan groups GP30, GP36, and GP38 showed the largest differences (Supplementary Fig. 1), each of them containing antennary fucosylated glycan (Fig. 2A).

Similarly, hs-CRP was lower in subjects with likely damaging HNF1A alleles than in those without HNF1A variants (0.21 [0.07–0.68] vs. 1.70 [0.60–3.91] mg/L; P = 3.09 ×10⁻⁵) (Supplementary Fig. 1).

Secondly, we examined whether GP30 and hs-CRP could serve as markers of HNF1A allele function. Both biomarkers were significantly lower in probands with likely damaging HNF1A alleles (n = 16) than in subjects with likely benign HNF1A alleles (n = 7); median GP30 0.43% (0.34–0.57) vs. 0.95% (0.51–1.79) (P = 0.012) and median hs-CRP 0.21 (0.07–0.68) vs 1.01 (0.84–2.36) mg/L ( P = 0.006) (Fig. 2B–E). Median GP30 and hs-CRP in subjects with VUS did not significantly differ from individuals with likely benign or without rare HNF1A alleles.

Examination of the classification performance of plasma N-glycans and hs-CRP was performed by ROC curve analysis, comparing biomarker values in subjects with likely damaging HNF1A alleles against subjects without rare HNF1A alleles (Fig. 2B–E).

Firstly, the discriminative performance of individual glycan groups was tested, where GP30, GP36, and GP38 showed the best discriminative power among all individual glycans, with AUCs of 0.90, 0.87, and 0.90, respectively (Fig. 2B–D). Secondly, a model based on the total plasma N-glycome (all 42 glycan groups included) was built. It showed a similar discriminative power between early-onset nonautoimmune diabetes and subjects with damaging HNF1A alleles when compared with the GP30, with AUCs for the total glycome model of 0.92 (95% CI 0.86−0.99) vs. 0.90 (95% CI 0.83−0.97) for GP30. hs-CRP also showed a satisfactory performance in distinguishing two groups, with an AUC of 0.83 (95% CI 0.71–0.94) (Fig. 2E). Finally, the joint performance of both GP30 and hs-CRP was calculated and resulted in an AUC of 0.90 (95% CI 0.83–0.98), which was again similar to GP30 alone.

The clinical potential of the best performing glycan, GP30, and hs-CRP was further evaluated. ROC curve analysis indicated that a diagnostic threshold of 0.70% for GP30 provided optimal discrimination of subjects with likely damaging HNF1A alleles from subjects with early-onset nonautoimmune diabetes and without HNF1A variants, showing sensitivity of 88% and specificity of 80%. A GP30 cutoff of 0.70% missed only 2 of 16 probands with likely damaging HNF1A alleles. If GP30 was used as a selective tool for stratification of the current cohort, 214 individuals (22%) with young-onset nonautoimmune diabetes would have HNF1A sequenced.

ROC curve analysis for hs-CRP showed that a threshold of 0.81 mg/L provided optimal discrimination between the groups, with 88% sensitivity and 69% specificity. Using the proposed cutoff, 2 of 16 subjects with likely damaging HNF1A alleles were also missed, whereas using it as a screening tool would have resulted in HNF1A sequencing of 269 individuals (27%) from this study.

In contrast, if we used classical clinical criteria for MODY genetic testing (diagnosis of diabetes <25 years of age, a family history of diabetes in at least two generations, endogenous insulin production, and negative GADAs), we would have picked up only 8 of 16 individuals (50%) with likely damaging HNF1A alleles while sequencing 99 individuals (10%) from this study.

The Exeter MODY probability calculator gives a pretest probability of any form of MODY but is not validated in subjects diagnosed with diabetes at >35 years (5) or of nonwhite ethnicity. In this study, 370 of the 989 participants could be assessed using the MODY calculator, and 136 (37% of those assessed) had an estimated probability of MODY of >20%. The MODY calculator could be used to assess 14 of 16 of the probands with damaging MODY variants, and 9 of these had an estimated probability of >20%, thus a sensitivity of 56% for detecting HNF1A-MODY (Table 2). This is similar to the classic criteria and also to performance in the Using pharmacogeNetics to Improve Treatment in Early-onset Diabetes (UNITED) study, where 55% of cases were missed by the calculator (<25% risk) (35). The calculator would also lead to selection of a higher proportion of the cases (37% compared with 22–27%).

We also examined GP30 and hs-CRP levels in subjects with VUS and novel HNF1A alleles to estimate whether biomarkers assisted in assigning the functional effect of the allele. Among the alleles labeled as VUS, the three individuals (a proband and two relatives) with variant p.P291T, a participant with variant p.T515M, and a participant with novel variant p.S3C had GP30 and hs-CRP above the proposed cutoffs, providing support for their benign effect. In contrast, the three subjects with variant p.A251T (a proband and two relatives) had hs-CRP below 0.30 mg/L; however, two had GP30 above and one had GP30 slightly below the cutoff value, making the results inconclusive. Regarding the remaining two novel HNF1A alleles, using GP30 and hs-CRP classified p.K222N as damaging (opposite to the functional work results), whereas p.G151S had discordant biomarker results.

In this study, we found 25 rare HNF1A alleles in 29 individuals (2.9% of the participants). After the systematic assessment of these alleles, we considered that 12 HNF1A alleles (in 16 probands) are likely to be damaging HNF1A protein function, 9 are likely to be benign, and 4 remain as variants of unknown significance.

The participants in this study were all found to have HNF1A-MODY as a result of participating in clinical research, demonstrating that many cases are missed in real-life clinical practice. The consequence of the diagnosis was that 10 individuals from the 16 probands and 2 relatives were able to commence sulfonylurea treatment and 4 discontinued insulin treatment.

This study showed that antennary fucosylated plasma glycans (GP30, GP36, and GP38) and hs-CRP levels were significantly lower in subjects harboring likely damaging HNF1A alleles compared with individuals without rare HNF1A alleles. This finding is consistent with the role of HNF1A, which acts as transcription factor for CRP (9) and genes encoding fucosyltransferases (13) in hepatocytes. Results of this study also confirmed our previous findings where we examined the plasma N-glycans and hs-CRP in groups of individuals already known to have MODY (12,17). However, here we examined the performance of the biomarkers in a relatively unselected population of subjects with young adult–onset nonautoimmune diabetes, which better reflects the situation encountered by clinicians while assigning a diagnosis.

The biomarkers GP30 and hs-CRP were equally successful in recognizing subjects with likely deleterious HNF1A alleles (88% sensitivity); however, GP30 showed better specificity than hs-CRP (80% vs. 69%). Moreover, compared with both classic clinical criteria and the MODY probability calculator, both biomarkers were superior in selecting subjects to be referred for genetic testing.

Thus, incorporating biomarkers into clinical use of such prediction models may assist the successful stratification of individuals with young adult–onset diabetes carrying potentially deleterious HNF1A alleles.

In many countries, panel testing of many genes for monogenic diabetes is the first-line genetic test. Biomarkers may still be useful as an estimate of pretest risk of HNF1A-MODY in the context of interpreting panel results. In other counties, including Croatia currently, very little genetic testing is routinely available, so the biomarkers may continue to have a role in identifying those at greatest risk of HNF1A-MODY for single gene Sanger sequencing approaches.

Large-scale sequencing studies in both healthy and disease populations have shown that many variants initially thought to be disease causing are present in population samples in frequencies greater than would be expected for a rare monogenic condition (36). GP30 and hs-CRP could provide an additional value in assigning disease causality of identified HNF1A alleles, because individuals with likely damaging HNF1A alleles had significantly lower levels of antennary fucosylated glycans and hs-CRP than subjects with benign HNF1A alleles. Both biomarkers were consistent in assigning a direction of the functional effect in eight individuals harboring VUS in this study (four probands and four relatives), placing the variant p.A251T as a damaging one and p.S3C, p.P291T, and p.T515M as likely benign ones. The biomarkers are likely to be particularly useful in assessment of variants where there is most doubt over the functional consequences; for example, novel missense variants, which have been found in population sequencing databases at an allele frequency of ≤0.005%. It seems likely that most variants with a higher percentage MAF will be benign.

Since initiating this study, it has become clear that the phenotypical spectrum of HNF1A alleles is much wider than originally thought, so that although functionally deleterious alleles frequently cause MODY, this can by no means be assumed for every individual who possesses that allele (36). This shows the complexity of the interpretation of genetic variation, which we consider a limitation of our or any similar studies. In clinical practice, every case needs to be assessed on the basis of the individual phenotype and the predicted functional effect of the HNF1A allele. Cosegregation of allele with diabetes in a family and clinical response to sulfonylureas will help confirm the diagnosis.

In conclusion, we found that the biomarkers GP30 and hs-CRP could differentiate individuals with early-onset diabetes and likely damaging HNF1A alleles from those with young adult–onset nonautoimmune diabetes and without rare HNF1A alleles. A diagnostic protocol combining clinical features with biomarkers could improve the selection of subjects for genetic testing for HNF1A-MODY, the commonest form of monogenic diabetes in adults. Currently, easier availability of the hs-CRP assay makes it a more immediate prospect, whereas for wider use of N-glycans, a simpler assay for determining antennary fucose levels would have to be developed.

Acknowledgments. The authors thank all of the participants, research nurses, local investigators, and other clinical staff for their contribution to this study. The authors also thank Dr. Beverley Shields at the University of Exeter for providing the MODY probability model.

Funding. This work was funded by European Foundation for Study of Diabetes/Novo Nordisk Research Grant, European Community’s Seventh Framework Programme IntegraLife (contract number 315997), Hrvatska Zaklada za Znanost (Croatian Science Foundation) (UIP-2014-09-7769), Diabetes UK (reference number 14/0004735), and supported by the National Institute for Health Research Oxford Biomedical Research Centre. A.L.G. is a Wellcome Trust Senior Fellow in Basic Biomedical Science (095101/Z/10/Z and 200837/Z/16/Z). A.J. is a Diabetes UK George Alberti fellow.

Duality of Interest. G.L. declares that he is a founder and owner and F.V. declares that he is an employee of Genos Ltd., which offers commercial service of glycomic analysis and has several patents in this field. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.J., T.P., F.V., A.J.B., N.S., E.P.M., C.J.G., M.Š., K.C., C.B., N.R.P., M.V.L., J.Ć.K., T.J.J., A.L.G., G.L., M.I.M., K.R.O., and O.G. read, critically revised, and approved the final manuscript. A.J., T.P., F.V., K.R.O., and O.G. analyzed and interpreted data. A.J., T.P., A.J.B., N.S., C.J.G., M.Š., K.C., C.B., M.V.L., J.Ć.K., and T.J.J. acquired data. A.J., T.P., K.R.O., and O.G. wrote the manuscript. E.P.M., N.R.P., and K.R.O. enrolled patients. A.L.G., G.L., M.I.M., K.R.O., and O.G. designed the study. O.G. and K.R.O. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented in abstract form at the Annual Meeting of the Society for Glycobiology, San Francisco, CA, 1–4 December 2015; at the 23rd International Symposium on Glycoconjugates, Split, Croatia, 15–20 September 2015; at the 6th Meeting of the EASD Study Group on Genetics of Diabetes, 11–13 May 2017, Leiden, the Netherlands; and at the 53rd Annual Meeting of the EASD, Lisbon, Portugal, 11–15 September 2017.