Glucose and paracrine regulation of α-cells, particularly with respect to sex differences, is still poorly understood. Here, we imaged islets from GluCre:GCaMP6f mice in pancreatic slices, additionally loaded with a red Ca²⁺ indicator, to precisely interrogate Ca²⁺ dynamics in α-cells and the adjacent β- and δ-cells. During a glucose ramp (1.8-10.8 mM), α-cell Ca²⁺ oscillations were heterogeneous, antiphasic to β-cells, and inversely correlated with δ-cells. Selective inhibition of insulin receptor (InsR) and somatostatin receptor subtypes-2 and 3 (SSTR2/3) shifted the majority of α-cells to hyperactivity, delayed the decline of α-cell activity and onset of β-cell Ca²⁺ oscillations during the ramp. While female α-cells exhibited greater sensitivity to SSTR2/3 inhibition, male α-cells were more responsive to InsR blockade. Under complete SSTR2/3 and InsR antagonism, α-cells exhibited elevated Ca²⁺ oscillations but remained glucose-dependent. We conclude that intra-islet coordination fine-tunes α-cell glucose responses, with sex-specific paracrine signaling potentially shaping glucagon profiles in metabolic disease.

Glucagon-like peptide-1 receptor agonists (GLP-1RAs) are effective therapies for type 2 diabetes (T2D) and obesity, yet patient responses are variable, with GLP1R gene variation potentially linked to therapeutic outcomes. A GLP1R natural missense variant, A316T, protects against T2D and cardiovascular disease. Here, we generated and characterized a human GLP1R A316T mouse model. Human GLP1RA316T/A316T mice displayed lower fasting blood glucose versus wild-type littermates even under metabolic stress, as well as slower weight gain and alterations in islet cytoarchitecture, glucagon secretion, and liver metabolism under a high-fat, high-sucrose diet. This was however associated with blunted responses to pharmacological GLP-1RAs in vivo. Further investigations in β cell models demonstrated that human GLP1R A316T exhibits characteristics of constitutive activation but dampened GLP-1RA responses. Results are further supported by cryo-EM analyses and molecular dynamics simulations of GLP-1R A316T structure, collectively demonstrating that the A316T variant governs basal GLP-1R activity and pharmacological responses to GLP-1R-targeting therapies.

Despite recent therapeutic advances, achieving optimal glycaemic control remains a challenge in managing type 2 diabetes. Sodium–glucose cotransporter 2 (SGLT2) inhibitors have emerged as effective treatments by promoting urinary glucose excretion. However, the full scope of their mechanisms extends beyond glycaemic control. At present, their immunometabolic effects remain elusive. To investigate the effects of SGLT2 inhibition or deletion, we compared the metabolic and immune phenotype between high-fat-diet-fed control mice, mice treated chronically with dapagliflozin, and total-body Slc5a2-knockout mice. SGLT2-null mice exhibited better glucose tolerance and insulin sensitivity (blood glucose during IPGTT AUC 0–90 min 1175 ± 57.4 mmol/l × min, mean ± SEM) compared with control (AUC 0–90 min 1857 ± 117.9 mmol/l × min, p=0.05) or dapagliflozin-treated mice (AUC 0–90 min 1506 ± 68.72 mmol/l × min, p=0.09), independent of glycosuria and body weight. Moreover, SGLT2-null mice demonstrated physiological regulation of corticosterone secretion, with lower morning levels than control mice (p<0.01). Systemic cytokine profiling also unveiled significant alterations in inflammatory mediators, particularly IL-6. Furthermore, unbiased proteomic analysis demonstrated downregulation of acute-phase proteins and upregulation of glutathione-related proteins, suggesting a role in the modulation of antioxidant responses. Conversely, IL-6 treatment increased SGLT2 expression in human kidney HK2 cells, suggesting a role for cytokines in the effects of hyperglycaemia. Collectively, our data elucidate a potential interplay between SGLT2 activity, immune modulation and metabolic homeostasis, as well as a potential feedback loop between SGLT2 expression and cytokine concentration.

Defective insulin secretion is a hallmark of diabetes mellitus. Glucose-induced Ca2+ oscillations are critical for the stimulation of insulin secretion, though the mechanisms through which these propagate across the islet are poorly understood. Here, we use beta cell-targeted GCaMP6f to explore the role of endoplasmic reticulum (ER) Ca2+ mobilization in response to submaximal (11mM) or hyperglycemic (25mM) glucose, mimicking diabetes. Inhibition of inositol 1,4,5-trisphosphate (IP3) receptors, and other ion channels, with 2-aminoethoxydiphenyl borate (2-APB), had minimal effects on the initial peak or intercellular connectivity provoked by 11mM glucose. However, 2-APB lowered subsequent glucose-induced cytosolic Ca2+ increases and connectivity at both 11 and 25mM glucose. Unexpectedly, the activation of IP3 receptors with the muscarinic acetylcholine receptor agonist carbachol had minimal impact on the initial peak elicited by 11 mM glucose, but Ca2+ waves at 11 and 25 mM glucose were more poorly coordinated. To determine whether ER calcium mobilization was sufficient to initiate Ca2+ waves we next blocked sarco(endo)plasmic Ca2+ ATPase (SERCA) pumps with thapsigargin, whilst preventing plasma membrane depolarization with the KATP-channel opener, diazoxide. Under these conditions, an initial cytosolic Ca2+ increase was followed by secondary Ca2+ waves that subsided slowly. The application of carbachol alongside diazoxide still enhanced Ca2+ dynamics, though activity was uncoordinated. After genetic deletion of SERCA2 in beta cells, Ca2+ wave frequency, but not connectivity, were lowered. Our results show that ER Ca2+ mobilization plays a relatively minor role in the initiation and propagation of Ca2+ waves in response to glucose but is needed for sustained Ca2+ waves.

Aims/Hypothesis:The importance for normal insulin secretion of ceramide synthesis is unclear. De novo ceramide synthesis requires serine palmitoyl transferase, SPT2, encoded by Sptl2. Methods:We generated β-cell-selective Sptl2 null mice by crossing animals with floxed alleles to mice expressing Cre recombinase from the Ins1 locus. Metabolic phenotyping, transcriptomic, functional analyses and histology were performed using standard approaches. Results:Islets from Sptlc2 ΔInsl mice displayed marked alterations in ceramide and sphingomyelin levels: ceramide content: p=0.016 and p=0.109; sphingomyelin content: p=0.016 and p=0.004 in Sptlc2 ΔInsl vs Sptlc2 CTL mice under regular and high fat diet, respectively, despite compensatory increases in the expression of enzymes in the salvage and sphingomyelinase pathways. Correspondingly, profound abnormalities were observed in glucose-regulated insulin secretion and glucose tolerance in vivo, both on a regular chow and high fat diet. These changes were associated with a drastic (~80%) lowering in β-cell numbers, and a more minor increase in delta cell numbers. They were also preserved in animals maintained on a ketogenic diet, consistent with a cell autonomous effect on the β-cell. Despite normal glucose-regulated intracellular calcium dynamics and insulin secretion, marked transcriptomic changes were observed in Sptlc2 ΔInsl mouse islets, with affected GO terms including lysosome organisation and regulation of autophagy. Consistent with roles for compromised SPT2 function in diseased β-cells, Sptl2 expression in Balbc and DBA2J mouse islets was lowered by a high fat-diet. Moreover, SPTLC2 mRNA tended to be lower, and SPTLC1 mRNA was significantly decreased, in islets from human subjects with type 2 diabetes versus normoglycemic individuals. Conclusions:Preserved de novo ceramide synthesis is required to maintain normal β-cell mass and thus insulin secretion in mice. Therapeutic approaches which seek to target this process systemically using pharmacological SPT2 inhibitors should thus be treated with caution.

Introduction and Objective: GLP1 and other incretin receptor agonists are used extensively in the clinic to treat type 2 diabetes. Little is known currently about their effect on islet-wide β-cell calcium dynamics or connectivity in vivo. Methods: We engrafted islets expressing the Ca2+ indicator GCaMP6 into the anterior chamber of the eye (ACE) of control, high fat high sucrose (HFHS) diet- or genetically-induced hyperglycemic mice. Following engraftment and diet treatment (21 weeks), islet-wide Ca2+ dynamics were recorded in 2D and 3D by confocal imaging through the cornea of anesthetized animals at day 0, 1 and 7 (n=9 islets, 6 mice). Results: In control animals, Ca2+ waves routinely emanated from “leader” β-cells at the islet periphery. Islets engrafted into HFHS diet animals or mice heterozygous for a hypomorphic Gck allele (<50% residual expression; C57BL6/J-Gcktm1(mCard)/Rutt; GckKI+, Salazar, S. et al, Biorxiv, 2024), displayed lower connectivity (0.89 in control vs 0.61 HFHS vs 0.59 GckKI+, respectively). Acute intraperitoneal exendin-4 (Ex-4, 1 nmol/kg) administration at day 8 restored islet Ca2+ dynamics in HFHS and GckKI+ animals (Ctrl+PBS vs Ctrl+Ex-4 not significant, n=4, HFD+PBS vs HFD+Ext-4 p=0.013, n=7, GckKI++PBS vs GckKI++Ext-4, p=0.016, n=5). Notably, Ex-4 reengaged a “hub” population, restoring β-cell-β-cell connectivity and Ca2+ dynamics Conclusion: An effect of hyperglycemia to disconnect islet-wide functional networks in vivo is rapidly reversed by GLP1R agonism, likely contributing to improved insulin secretion. The actions of Ex-4 were more marked than usually observed in vitro, possibly implicating neuron-mediated signal relay to the β-cell from remotely-located GLP1R (e.g. in the gut, brain). Further studies will be required to explore this possibility. L.F. Delgadillo Silva: None. S. Larouche: None. A. Prat: None. S. Salazar: None. G.A. Rutter: Consultant; Sun Pharmaceutical Industries Ltd. CIHR/IRSC Post-doctoral Fellowship (489982)

The bromodomain and extra-terminal (BET) protein family acts as ‘epigenetic readers’ to identify the acetylation marks on histones that convert the acetylated lysine residues into observable phenotypes. BET proteins have gained attention due to their ability to modulate the transcription of pathology-related genes involved in cancer and autoimmune diseases, including type 1 diabetes mellitus. However, targeting BET proteins may have secondary effects on other host cells. We aimed to elucidate possible secondary effects of BET inhibition on pancreatic beta cell function. We studied the effect of the small-molecule BET inhibitor I-BET151 on pancreatic beta cells in vitro, ex vivo and in vivo. GTTs, ITTs and glucose-stimulated insulin secretion assays were performed in healthy mice and a mouse model of diabetes following daily i.p. injections of I-BET151 for 2 weeks. Transcriptomic analysis was carried out on primary mouse islets, which were subjected to ex vivo I-BET151 treatment. Changes in expression were further validated in primary human islets. Administration of I-BET151 modestly but significantly increased glucose excursions and reduced insulin responses in both healthy mice and diabetic mice. We found that I-BET151 exposure significantly reduced the expression of Hnf4α (also known as Hnf4a; MODY1), Gck (MODY2), Hnf1α (also known as Hnf1a; MODY3), Glut2 and other genes essential for beta cell function in rat INS-1E insulinoma cells and in mouse primary islets and human islets. Global gene expression analysis in cells treated with I-BET151 showed a downregulation of the phosphoinositide-3-kinase (PI3K)–Akt pathway. Downregulation of forkhead box protein O1, a downstream transcriptional factor of the PI3K–Akt pathway, partially rescued I-BET151-driven downregulation of Gck and insulin secretion. Likewise, islets from I-BET151-treated mice showed a modest reduction in glucose-stimulated insulin secretion. The results presented here suggest that BET inhibition therapy should be used with caution due to possible bimodal effects at high concentrations at the detriment of pancreatic beta cell function.

The coordinated function of beta cells within the pancreatic islet is required for the normal regulation of insulin secretion and is partly controlled by specialized "leader" and highly connected "hub" beta-cell subpopulations. Whether cells within these subpopulations are functionally stable in vivo remains unclear. Here, we establish an approach to monitor Ca 2+ dynamics within individual beta cells over time, after engraftment into the anterior eye chamber, where continuous blood perfusion and near normal innervation pertain. Under normoglycemic conditions, islet network dynamics, and the behavior of individual leaders and hubs, remain stable for at least seven days. Hyperglycemia, resulting from high-fat diet feeding or the loss of a host Gck allele, caused engrafted islets to display incomplete and abortive Ca 2+ waves and overall connectivity was diminished. Whereas hub cell numbers were lowered profoundly in both disease models, leaders largely persisted. Treatment with the GLP1R agonist Exendin-4 led to a recovery of islet-wide Ca 2+ dynamics and the re-emergence of hub cells within minutes, with the effects of the incretin mimetic being more marked than those observed after analogous treatments in vitro . Similar observations were made using 3-dimensional imaging across the whole islet. Our findings thus suggest that incretins may act both directly and indirectly on beta cells in vivo. The approach described may provide broad applicability to the exploration of individual cell function over time in the living animal.

AbstractThe use of incretin analogues has emerged in recent years as an effective approach to achieve both enhanced insulin secretion and weight loss in type 2 diabetes (T2D) patients. Agonists which bind and stimulate multiple receptors have shown particular promise. However, off target effects, including nausea and diarrhoea, remain a complication of using these agents, and modified versions with optimized pharmacological profiles and/or biased signaling at the cognate receptors are increasingly sought. Here, we describe the synthesis and properties of a molecule which binds to both glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide (GIP) receptors (GLP-1R and GIPR) to enhance insulin secretion. HISHS-2001 shows increased affinity at the GLP-1R, as well as a tendency towards reduced internalization and recycling at this receptorversusFDA-approved dual GLP-1R/GIPR agonist tirzepatide. HISHS-2001 also displayed significantly greater bias towards cAMP generationversusβ-arrestin 2 recruitment compared to tirzepatide. In contrast, Gαsrecruitment was lowerversustirzepatide at the GLP-1R, but higher at the GIPR. Administered to obese hyperglycaemicdb/dbmice, HISHS-2001 increased circulating insulin whilst lowering body weight and HbA1c with similar efficacy to tirzepatide at substantially lower doses. Thus, HISHS-2001 represents a novel dual receptor agonist with an improved pharmacological profile.

Long non-coding RNAs (lncRNAs) are emerging as crucial regulators of beta cell function. Here, we show that an lncRNA-transcribed antisense to Pax6, annotated as Pax6os1/PAX6-AS1, was upregulated by high glucose concentrations in human as well as murine beta cell lines and islets. Elevated expression was also observed in islets from mice on a high-fat diet and patients with type 2 diabetes. Silencing Pax6os1/PAX6AS1 in MIN6 or EndoC-bH1 cells increased several beta cell signature genes' expression. Pax6os1/PAX6AS1 was shown to bind to EIF3D, indicating a role in translation of specific mRNAs, as well as histones H3 and H4, suggesting a role in histone modifications. Important interspecies differences were found, with stronger phenotype in humans. Only female Pax6os1 null mice fed a high-fat diet showed slightly enhanced glucose clearance. In contrast, silencing PAX6-AS1 in human islets enhanced glucose-stimulated insulin secretion and increased calcium dynamics, whereas overexpression of the lncRNA resulted in the opposite phenotype.

Proteotoxicity is a contributor to the development of type 2 diabetes (T2D), but it is unknown whether protein misfolding in T2D is generalized or has special features. Here, we report a robust accumulation of misfolded proteins within the mitochondria of human pancreatic islets in T2D and elucidate its impact on β cell viability. Surprisingly, quantitative proteomics studies of protein aggregates reveal that human islets from donors with T2D have a signature more closely resembling mitochondrial rather than ER protein misfolding. The matrix protease LonP1 and its chaperone partner mtHSP70 were among the proteins enriched in protein aggregates. Deletion of LONP1 in mice yields mitochondrial protein misfolding and reduced respiratory function, ultimately leading to β cell apoptosis and hyperglycemia. Intriguingly, LONP1 gain of function ameliorates mitochondrial protein misfolding and restores human β cell survival following glucolipotoxicity via a protease-independent effect requiring LONP1-mtHSP70 chaperone activity. Thus, LONP1 promotes β cell survival and prevents hyperglycemia by facilitating mitochondrial protein folding. These observations may open novel insights into the nature of impaired proteostasis on β cell loss in the pathogenesis of T2D that could be considered as future therapeutic targets.

AIMS:The use of incretin analogues has emerged as an effective approach to achieve both enhanced insulin secretion and weight loss in Type 2 diabetes (T2D) patients. Agonists which bind and stimulate multiple receptors have shown particular promise. However, off-target effects remain a complication of using these agents, and modified versions with optimised pharmacological profiles and/or biased signalling are sought. MATERIALS AND METHODS:Ligand synthesis was achieved using standard solid-phase techniques. Assessments of GLP-1R-binding kinetics, G protein recruitment and receptor internalisation were performed using biochemical and imaging approaches. Insulin secretion was measured in purified mouse and human islets, and drug efficacy was assessed in hyperglycaemic db/db mice. RESULTS:We describe the synthesis and properties of a molecule which binds to both glucagon-like peptide-1 (GLP-1) and glucose-dependent insulinotropic polypeptide (GIP) receptors (GLP-1R and GIPR) to enhance insulin secretion. HISHS-2001 shows increased affinity at the GLP-1R, as well as a tendency towards reduced internalisation and recycling at this receptor versus FDA-approved dual GLP-1R/GIPR agonist tirzepatide. HISHS-2001 also displayed significantly greater bias towards cAMP generation versus β-arrestin 2 recruitment compared to tirzepatide. In contrast, Gαs recruitment was lower versus tirzepatide at the GLP-1R, but unchanged at the GIPR. Administered to obese hyperglycaemic db/db mice, HISHS-2001 increased circulating insulin whilst lowering body weight and HbA1c with similar efficacy to tirzepatide at substantially lower doses. CONCLUSION:HISHS-2001 represents a novel dual receptor agonist with a promising pharmacological profile and actions. Future clinical studies will be needed to assess the safety and efficacy of this molecule in humans.

Monogenic diabetes, formerly called Maturity-Onset Diabetes of the Young (MODY), involves single-gene mutations, typically with dominant inheritance, and has been associated with variants in 14 genes. Among these, HNF1A mutations are the most common, and their diagnosis allows the use of alternative therapies, including sulfonylureas. In an earlier study, we described a variant displaying recessive transmission, p.A251T (Misra, S et al, Diabetes Care, 2020). Initial functional studies revealed only a modest impact on protein function. We extend these earlier in vitro studies to demonstrate that beta-like cells derived from pluripotent stem cells from variant carriers show impaired differentiation into insulin-positive cells, whereas differentiation into alpha cells is significantly enhanced. Additionally, mutant cells showed impaired glucose-stimulated insulin secretion but partially preserved responsiveness to treatment with sulfonylureas. Our study provides proof of principle for the utility of using patient-derived stem cells as a platform to assess the pathogenicity of HNF1A variants, and to explore potential treatment strategies.

Using 13C6 glucose labeling coupled to gas chromatography-mass spectrometry and 2D 1H-13C heteronuclear single quantum coherence NMR spectroscopy, we have obtained a comparative high-resolution map of glucose fate underpinning β cell function. In both mouse and human islets, the contribution of glucose to the tricarboxylic acid (TCA) cycle is similar. Pyruvate fueling of the TCA cycle is primarily mediated by the activity of pyruvate dehydrogenase, with lower flux through pyruvate carboxylase. While the conversion of pyruvate to lactate by lactate dehydrogenase (LDH) can be detected in islets of both species, lactate accumulation is 6-fold higher in human islets. Human islets express LDH, with low-moderate LDHA expression and β cell-specific LDHB expression. LDHB inhibition amplifies LDHA-dependent lactate generation in mouse and human β cells and increases basal insulin release. Lastly, cis-instrument Mendelian randomization shows that low LDHB expression levels correlate with elevated fasting insulin in humans. Thus, LDHB limits lactate generation in β cells to maintain appropriate insulin release.

Glucokinase (GK) catalyses the key regulatory step in glucose-stimulated insulin secretion. Correspondingly, hetero- and homozygous mutations in human GCK cause maturity-onset diabetes of the young (GCK-MODY) and permanent neonatal diabetes (PNDM), respectively. To explore the possible utility of glucokinase activators (GKA) and of glucagon-like receptor-1 (GLP-1) agonists in these diseases, we have developed a novel hypomorphic Gck allele in mice encoding an aberrantly spliced mRNA deleted for exons 2 and 3. In islets from homozygous knock-in (GckKI/KI) mice, GK immunoreactivity was reduced by >85%, and glucose-stimulated insulin secretion eliminated. Homozygous GckKI/KI mice were smaller than wildtype littermates and displayed frank diabetes (fasting blood glucose >18 mmol/L; HbA1c ~12%), ketosis and nephropathy. Heterozygous GckKI/+ mice were glucose intolerant (HbA1c ~5.5%). Abnormal glucose-stimulated Ca2+ dynamics and beta cell-beta cell connectivity in GckKI/+ islets were completely reversed by the recently-developed GKA, dorzagliatin, which was largely inactive in homozygous GckKI/KI mouse islets. The GLP-1 receptor agonist exendin-4 improved glucose tolerance in male GckKI/+ mice, an action potentiated by dorzagliatin, in male but not female mice. Sex-dependent additive effects of these agents were also observed on insulin secretion in vitro. Combined treatment with GKA and incretin may thus be useful in GCK-MODY or GCK-PNDM.

Variants at the SLC30A8 locus are associated with type 2 diabetes (T2D) risk. The lead variant, rs13266634, encodes an amino acid change, Arg325Trp (R325W), at the C-terminus of the secretory granule-enriched zinc transporter, ZnT8. Although this protein-coding variant was previously thought to be the sole driver of T2D risk at this locus, recent studies have provided evidence for lowered expression of SLC30A8 mRNA in protective allele carriers. In the present study, combined allele-specific expression (cASE) analysis in human islets revealed multiple variants that influence SLC30A8 expression. Epigenomic mapping identified an islet-selective enhancer cluster at the SLC30A8 locus, hosting multiple T2D risk and cASE associations, which is spatially associated with the SLC30A8 promoter and additional neighbouring genes. Deletions of variant-bearing enhancer regions using CRISPR-Cas9 in human-derived EndoC-βH3 cells lowered the expression of SLC30A8 and several neighbouring genes, and improved insulin secretion. Whilst down-regulation of SLC30A8 had no effect on beta cell survival, loss of UTP23, RAD21 or MED30 markedly reduced cell viability. Although eQTL or cASE analyses in human islets did not support the association between these additional genes and diabetes risk, the transcriptional regulator JQ1 lowered the expression of multiple genes at the SLC30A8 locus and enhanced stimulated insulin secretion.

Introduction & Objective: GL0034 (GL), a once weekly glucagon-like peptide 1 receptor agonist, previously demonstrated significant reductions in body weight (BW) up to Day 22 in a single ascending dose study in individuals with obesity. This phase 1 study assessed the safety, tolerability and metabolic effects of GL after multiple ascending doses. Methods: Individuals with BMI ≥28 kg/m2 (N=24) were randomized (9:3) to subcutaneous GL, fixed doses (4 × 680 µg; cohort 1); or increasing doses (680, 900, 1520, 2000 µg; cohort 2) or placebo, once weekly for four weeks. Safety, tolerability and key metabolic parameters were assessed. Results: Most common adverse events (AE) were gastrointestinal (GI) with dose-dependent nausea, decreased appetite and vomiting. One individual with a GI related serious AE rapidly recovered upon treatment with intravenous rehydration. On Day 23, reduction was observed in all parameters from baseline (BL) with significant reductions in glucose area under the curve and HbA1c in both groups. In cohorts 1 & 2, BW reduction versus BL was 2.9 kg and 4.6 kg respectively on Day 29 (Table). Conclusions: In individuals with obesity, once weekly GL dosing for four weeks, demonstrated clinically relevant reductions in glucose, insulin, HbA1c, lipids and BW with an overall good tolerability. Disclosure R. Thennati: None. V.S. Burade: None. M. Natarajan: None. P. Shahi: None. R. Nagaraja: None. S.K. Agrawal: None. T. Duvauchelle: Consultant; Sun Pharmaceutical Industries Ltd. A. Garcia-Ocana: Consultant; Sun Pharmaceutical Industries Ltd. G.A. Rutter: Advisory Panel; Sun Pharmaceutical Industries Ltd. R.E. Pratley: Other Relationship; Bayer AG, Dompé, Endogenex, Inc., Gasherbrum Bio, Inc., Hengrui (USA) Ltd., Intas Pharmaceuticals Ltd., Eli Lilly and Company, Merck Sharp & Dohme Corp., Novo Nordisk, Novo Nordisk, Pfizer Inc., Rivus Pharmaceuticals Inc., Sun Pharmaceutical Industries Ltd. Consultant; AbbVie Inc., AstraZeneca. Other Relationship; Bayer HealthCare Pharmaceuticals, Inc., Biomea Fusion, Carmot Therapeutics, Inc., Corcept Therapeutics, Fractyl Health, Inc., Genprex. Consultant; Getz Pharma. Other Relationship; Lilly USA LLC, Sanofi. Consultant; Scholar Rock, Inc. B. Thorens: Advisory Panel; Sun Pharmaceutical Industries Ltd. T. Vilsbøll: Consultant; AstraZeneca. Advisory Panel; Boehringer-Ingelheim. Speaker's Bureau; Mundipharma. Advisory Panel; Novo Nordisk, Lilly Diabetes, Sanofi. Speaker's Bureau; Bayer Inc., Gilead Sciences, Inc. Advisory Panel; Sun Pharmaceutical Industries Ltd. Research Support; Lilly Diabetes.

Aims/hypothesis People with type 2 diabetes are heterogeneous in their disease trajectory, with some progressing more quickly to insulin initiation than others. Although classical biomarkers such as age, HbA 1c and diabetes duration are associated with glycaemic progression, it is unclear how well such variables predict insulin initiation or requirement and whether newly identified markers have added predictive value. Methods In two prospective cohort studies as part of IMI-RHAPSODY, we investigated whether clinical variables and three types of molecular markers (metabolites, lipids, proteins) can predict time to insulin requirement using different machine learning approaches (lasso, ridge, GRridge, random forest). Clinical variables included age, sex, HbA 1c , HDL-cholesterol and C-peptide. Models were run with unpenalised clinical variables (i.e. always included in the model without weights) or penalised clinical variables, or without clinical variables. Model development was performed in one cohort and the model was applied in a second cohort. Model performance was evaluated using Harrel’s C statistic. Results Of the 585 individuals from the Hoorn Diabetes Care System (DCS) cohort, 69 required insulin during follow-up (1.0–11.4 years); of the 571 individuals in the Genetics of Diabetes Audit and Research in Tayside Scotland (GoDARTS) cohort, 175 required insulin during follow-up (0.3–11.8 years). Overall, the clinical variables and proteins were selected in the different models most often, followed by the metabolites. The most frequently selected clinical variables were HbA 1c (18 of the 36 models, 50%), age (15 models, 41.2%) and C-peptide (15 models, 41.2%). Base models (age, sex, BMI, HbA 1c ) including only clinical variables performed moderately in both the DCS discovery cohort (C statistic 0.71 [95% CI 0.64, 0.79]) and the GoDARTS replication cohort (C 0.71 [95% CI 0.69, 0.75]). A more extensive model including HDL-cholesterol and C-peptide performed better in both cohorts (DCS, C 0.74 [95% CI 0.67, 0.81]; GoDARTS, C 0.73 [95% CI 0.69, 0.77]). Two proteins, lactadherin and proto-oncogene tyrosine-protein kinase receptor, were most consistently selected and slightly improved model performance. Conclusions/interpretation Using machine learning approaches, we show that insulin requirement risk can be modestly well predicted by predominantly clinical variables. Inclusion of molecular markers improves the prognostic performance beyond that of clinical variables by up to 5%. Such prognostic models could be useful for identifying people with diabetes at high risk of progressing quickly to treatment intensification. Data availability Summary statistics of lipidomic, proteomic and metabolomic data are available from a Shiny dashboard at https://rhapdata-app.vital-it.ch . Graphical Abstract