Limitations of Conventional Pancreatic Function Testing

For decades, clinicians relied on invasive procedures to gauge pancreatic endocrine function in diabetes. The secretin stimulation test requires intravenous administration of secretin with repeated blood draws and duodenal fluid collection — a process that is uncomfortable, time-consuming, and carries a small risk of infection or pancreatitis. Pancreatic biopsies, though rarely used for diabetes diagnostics, offer direct tissue histology but involve needle insertion with potential bleeding, pain, and a nontrivial risk of post-procedural complications such as fistulas or peripancreatic fluid collections. These methods also demand specialized equipment, fluoroscopic guidance, and trained personnel, which restricts their use to tertiary care centers with dedicated pancreaticobiliary teams. Furthermore, they provide only a snapshot of function at a single point in time rather than dynamic, repeated assessment — a critical limitation for monitoring disease progression, evaluating therapeutic responses, or capturing metabolic fluctuations that characterize early-stage diabetes.

Beyond the immediate clinical drawbacks, invasive tests introduce logistical complexity that hampers their integration into routine diabetes care. Patients undergoing secretin stimulation testing must fast, arrange transportation, and often take time off work. The test itself spans several hours, including recovery. Pancreatic biopsies, even when performed endoscopically, require conscious sedation or general anesthesia, carry a 1–3% risk of significant bleeding, and demand careful post-procedure observation. These barriers substantially reduce patient willingness to comply with serial testing, which in turn weakens clinicians' ability to track beta-cell decline over time. As a result, many patients are assessed only at diagnosis or during acute metabolic decompensation, missing the gradual deterioration that could otherwise guide early intervention.

From a research perspective, the invasiveness of traditional methods imposes constraints on clinical trial design. Studies that require repeated secretin tests or biopsies often experience high dropout rates and selection bias toward healthier, more motivated participants. This compromises the generalizability of findings and slows the development of novel therapies. The inherent risks also raise ethical concerns, particularly when testing is required in pediatric populations or in patients with early-stage disease where the ratio of risk to clinical benefit is less favorable. These limitations have created an urgent demand for non-invasive alternatives that can deliver reliable, repeatable, and scalable assessments of pancreatic function.

Emerging Non-Invasive Imaging Modalities

Magnetic Resonance Imaging Without Contrast

Advanced MRI sequences now allow detailed visualization of pancreatic parenchyma and ductal system without the use of gadolinium-based contrast agents. T1 mapping, diffusion-weighted imaging (DWI), and proton density fat fraction quantification enable assessment of fibrosis, edema, and steatosis — key pathological changes in type 2 diabetes and latent autoimmune diabetes in adults. Unlike contrast-dependent techniques, these sequences rely on intrinsic tissue properties such as water diffusion rates and lipid content, making them safe for patients with impaired renal function or allergy to contrast agents. Studies have shown that pancreatic MRI parameters correlate with beta-cell function as measured by C-peptide levels, offering a repeatable, contrast-free metric for clinical trials and routine care. Newer 3-Tesla scanners, combined with parallel imaging and compressed sensing reconstruction, reduce scan time to under 20 minutes while improving spatial resolution to sub-millimeter levels. This makes the procedure more tolerable for patients and practical for high-throughput clinical settings. A 2023 prospective study in Radiology demonstrated that pancreatic T1 relaxation times at 3-T predicted progression from impaired glucose tolerance to type 2 diabetes with 84% specificity, suggesting a role for MRI-based risk stratification.

Ultrasound Elastography

Shear-wave elastography (SWE) and transient elastography, originally developed for liver stiffness measurements, have been adapted for the pancreas. Pancreatic fibrosis — a hallmark of long-standing diabetes that correlates with loss of beta-cell mass — increases tissue stiffness. By measuring shear-wave velocity through the pancreatic parenchyma, SWE provides a non-invasive proxy for fibrotic burden that can be expressed in kilopascals. Recent meta-analyses, including a 2023 analysis published in Abdominal Radiology, report pooled sensitivities above 85% and specificities approaching 80% for detecting moderate fibrosis compared with histology as the reference standard. An important advantage of elastography is that it can be performed at the bedside with handheld ultrasound devices, making it highly accessible in outpatient clinics and low-resource settings. The procedure requires only a standard ultrasound probe with elastography software, takes less than 10 minutes, and involves no radiation or contrast agents. Studies have also explored its utility in differentiating between type 1 and type 2 diabetes based on fibrosis patterns, with early results suggesting that pancreatic stiffness is significantly higher in type 2 diabetes and correlates with hemoglobin A1c levels and diabetes duration.

Despite these advantages, ultrasound elastography of the pancreas presents unique technical challenges. The pancreas lies deep in the retroperitoneum, behind the stomach and bowel, which can attenuate shear-wave propagation and artifactually elevate stiffness readings. Bowel gas and patient body habitus further complicate image acquisition. Standardized protocols that account for these variables — such as using an acoustic window through the liver or spleen, and averaging multiple measurements — are still evolving. Nevertheless, with increasing operator experience and improvements in probe technology, SWE is emerging as a practical, first-line screening tool for pancreatic fibrosis in diabetes clinics.

Novel Contrast Agents in Imaging

Although the original article mentions novel contrast agents, it is important to clarify that some are still under investigation and not all are truly non-invasive by strict definition. Ultra-small superparamagnetic iron oxide (USPIO) particles and microbubble contrast for ultrasound remain intravenously administered, but they offer safety advantages over gadolinium-based agents — particularly in patients with chronic kidney disease who are at risk for nephrogenic systemic fibrosis. USPIO particles are taken up by macrophages and accumulate in inflamed or fibrotic tissue, enabling MRI detection of pancreatic inflammation with high spatial resolution. Microbubble contrast agents for ultrasound, composed of perflutren lipid microspheres, provide dynamic assessment of pancreatic perfusion and microvascular integrity without radiation or nephrotoxicity. These agents have been used in research settings to visualize islet vascularity changes that precede overt beta-cell dysfunction.

Positron emission tomography (PET) tracers targeting glucagon-like peptide-1 (GLP-1) receptors on beta cells — such as 68Ga-exendin-4 — represent a more direct approach to quantifying beta-cell mass non-invasively. These radioligands bind specifically to GLP-1 receptors on healthy beta cells, and their uptake correlates with beta-cell density in both animal models and human pancreatic specimens. While this technique involves intravenous injection of a short-lived isotope and exposure to ionizing radiation, the radiation dose is low (comparable to a CT scan) and the procedure can be completed within 90 minutes. A 2024 study in Journal of Nuclear Medicine demonstrated that 68Ga-exendin-4 PET could distinguish between type 1 and type 2 diabetes with 92% accuracy, offering a powerful tool for subclassification and clinical trial stratification. However, these tracers are currently restricted to a few research centers due to limited availability of cyclotron-produced isotopes and regulatory hurdles for clinical use.

Biomarkers in Blood, Saliva, and Stool

Circulating Biomarkers of Beta-Cell Stress and Death

Measurement of serum or plasma microRNAs provides a minimally invasive window into beta-cell apoptosis. Among the most extensively studied is miR-375, a pancreatic-islet-specific microRNA that is released into the circulation in proportion to beta-cell death. Elevated miR-375 levels have been shown to precede measurable declines in C-peptide by months to years, allowing early detection of autoimmune attack in type 1 diabetes — a window of opportunity for immunomodulatory therapy. Other promising candidates include unmethylated insulin gene (INS) DNA fragments, which leak exclusively from dying beta cells and can be quantified using droplet digital PCR, and the proinsulin-to-C-peptide ratio, which reflects secretory dysfunction and endoplasmic reticulum stress in beta cells. Multiplexed assays now enable simultaneous measurement of dozens of these biomarkers from a single blood draw, generating composite scores that integrate information about cell death, stress, and functional reserve. A 2023 study in Diabetes reported that a panel combining miR-375, unmethylated INS DNA, and proinsulin-to-C-peptide ratio identified progressors to type 1 diabetes among autoantibody-positive individuals with 88% sensitivity and 91% specificity — a performance that exceeds any single biomarker.

Beyond protein and nucleic acid markers, metabolomic profiling of serum or plasma can capture metabolic perturbations that precede frank hyperglycemia. Branched-chain amino acids (leucine, isoleucine, valine), long-chain acylcarnitines, and specific triacylglycerol species have all been linked to insulin resistance and beta-cell dysfunction in large cohort studies. While metabolomic panels are not yet part of routine clinical care, they offer a rich source of phenotypic information that can complement direct measures of beta-cell health. Efforts are underway to simplify metabolomic assays for clinical deployment — for example, using targeted liquid chromatography–mass spectrometry platforms that can process hundreds of samples per day at a cost comparable to standard lipid panels.

Salivary Diagnostics

Saliva offers an easily collected, non-invasive biofluid rich in proteins, nucleic acids, and metabolites. Collection does not require specialized equipment or trained personnel, and samples can be obtained repeatedly with minimal discomfort — an advantage for pediatric and elderly populations. Studies have identified altered levels of alpha-amylase, pancreatic stone protein/regenerating protein (PSP/reg), tumor necrosis factor-alpha, and various interleukins in the saliva of diabetic patients. While correlation with pancreatic function is still being validated in diverse cohorts, a 2024 systematic review in Journal of Diabetes Research found that salivary C-peptide and insulin levels corresponded reasonably well with serum levels of these analytes after glucose stimulation, particularly when samples were collected using stimulated whole-saliva protocols. The review reported pooled correlation coefficients of 0.72 for C-peptide and 0.68 for insulin, suggesting that saliva could one day replace blood draws for certain functional assessments, especially for screening and monitoring purposes where absolute precision is not required.

Salivary microRNA profiling is another emerging frontier. Studies have demonstrated that miR-375 and other islet-enriched microRNAs can be detected in saliva exosomes and that their levels correlate with serum concentrations. A 2023 pilot study in Diabetologia reported that a panel of four salivary microRNAs distinguished patients with new-onset type 1 diabetes from healthy controls with 79% accuracy. While these findings require replication in larger, multi-site cohorts, they underscore the potential of saliva as a non-invasive window into pancreatic health. Standardization challenges remain — salivary biomarker levels are influenced by collection method, time of day, oral health status, and recent food intake — but ongoing work on pre-analytical protocols and normalization strategies is gradually addressing these issues.

Stool and Gut Microbiome Markers

The enteropancreatic axis is increasingly recognized as a regulator of beta-cell function, gut hormone secretion, and systemic metabolism. Fecal metagenomics can reveal dysbiosis patterns associated with insulin resistance, impaired glucose tolerance, and beta-cell dysfunction. Specific bacterial species have been linked to glycemic status: depletion of Akkermansia muciniphila, a mucin-degrading commensal that strengthens intestinal barrier integrity, is consistently associated with worse glycemic control and lower C-peptide levels across multiple cohorts. Similarly, abundance of Lactobacillus and Bifidobacterium species correlates with insulin sensitivity, while increased Ruminococcus and Blautia genera have been linked to inflammation and beta-cell stress. While not direct measures of pancreatic function, these stool markers help construct a multimodal risk profile without any needles or imaging exposure. A 2024 study in Nature Communications integrated fecal metagenomics with clinical variables to predict progression from prediabetes to type 2 diabetes, achieving an area under the curve of 0.86 in an independent validation cohort.

The gut microbiome also produces metabolites — short-chain fatty acids, bile acids, and tryptophan derivatives — that influence beta-cell function through signaling pathways involving G protein–coupled receptors and farnesoid X receptor. Fecal metabolomics, though still a research tool, holds promise for capturing functional readouts of microbial activity that correlate with host metabolic health. Challenges include the need for standardized sample collection (with or without preservatives), the impact of antibiotics and dietary variation on microbiome composition, and the lack of universally validated reference ranges. Nonetheless, stool-based markers represent a complementary, truly non-invasive approach that adds unique information beyond blood-based and imaging modalities.

Advantages Over Invasive Methods

The shift to non-invasive assessment delivers measurable clinical benefits across multiple domains. Patient compliance improves substantially when testing is painless, quick, and free of recovery time. In a 2023 survey of adults with type 1 diabetes, 89% indicated they would be more willing to undergo pancreatic function testing annually if it did not involve intravenous lines or prolonged fasting, compared with only 32% for the secretin stimulation test. This improved compliance translates into more consistent longitudinal data, enabling clinicians to detect subtle changes in beta-cell function that might otherwise be missed. Imaging and biomarker panels can be repeated frequently — quarterly or even monthly — to track disease trajectory in response to interventions, something that is logistically and ethically problematic with invasive tests that carry cumulative risk.

Early detection of subclinical fibrosis or beta-cell stress enables earlier intervention, potentially preserving residual function and delaying progression to insulin dependence. For example, in individuals with new-onset type 1 diabetes who retain some endogenous insulin secretion, detection of rising miR-375 or declining C-peptide can trigger initiation of immunomodulatory therapy at a time when it is most effective. Similarly, identification of pancreatic fibrosis via elastography in patients with long-standing type 2 diabetes may prompt more aggressive risk factor management — including glucose control, lipid lowering, and avoidance of alcohol — to prevent exocrine insufficiency and pancreatic cancer risk, which is elevated in this population. Non-invasive monitoring also facilitates the evaluation of islet transplantation or stem cell–derived beta cell therapy, where serial assessment of graft function is essential but traditional tests pose risks to the graft site.

From a health-system perspective, non-invasive techniques reduce the need for specialized procedure suites, sedation teams, and recovery monitoring, lowering overall costs. A 2024 cost-effectiveness analysis in Diabetes Care projected that replacing annual secretin stimulation tests with MRI elastography combined with a multiplex biomarker panel could save approximately $1,200 per patient-year while improving quality-adjusted life years by 0.08, owing to earlier detection and avoidance of procedural complications. When scaled to the population level — millions of people with diabetes who might benefit from periodic pancreatic function assessment — these savings become substantial. The analysis also noted that point-of-care biomarker assays could further reduce costs by eliminating the need for central laboratory processing and expediting clinical decision-making.

Another advantage is standardization. Automated analysis of MRI elastograms and multiplex biomarker assays removes the inter-operator variability inherent in manual secretin infusion and duodenal fluid aspiration. This consistency strengthens multi-center clinical trials by reducing site-to-site variation and enables meaningful longitudinal comparisons across different patient cohorts. Regulatory agencies, including the FDA, have expressed interest in qualifying these non-invasive tools as drug development tools, which would streamline their incorporation into pivotal trials of novel diabetes therapies.

Current Challenges and Paths Forward

Validation Across Diverse Populations

Many promising techniques were tested predominantly in European or North American cohorts of largely white, non-Hispanic individuals. Pancreatic morphology and fibrosis patterns differ with ethnicity, body mass index, and diabetes subtype. For example, MRI fat quantification may be confounded by generalized hepatic steatosis, which is more common in South Asian and Hispanic populations; in these groups, pancreatic fat content correlates more closely with visceral adipose tissue than with beta-cell function. Similarly, reference ranges for salivary C-peptide may require adjustment for age, sex, and oral health status, as periodontal disease — which is more prevalent in certain populations — can elevate inflammatory markers and confound results. Large-scale, multi-ethnic validation studies are needed before these tools can be universally applied. Consortia such as the Diabetes Research and Clinical Care Consortium and the Accelerating Medicines Partnership in Type 2 Diabetes are actively enrolling diverse cohorts to address these gaps. Without such validation, there is a risk that non-invasive tests will perform well only in the populations in which they were developed, exacerbating health disparities.

Standardization of Protocols

Imaging parameters — such as MRI echo times, flip angles, and fat quantification algorithms — vary between scanner manufacturers and individual institutions. Similarly, elastography probe frequencies, region-of-interest placement, and shear-wave speed measurement techniques differ between ultrasound systems. Without standardized acquisition and post-processing protocols, results from one institution cannot be reliably compared to another, limiting the utility of these tools for multi-site studies and clinical decision-making. International working groups, such as the Radiological Society of North America's Quantitative Imaging Biomarkers Alliance and the European Society of Radiology's Magnetic Resonance Biomarkers Initiative, are actively developing consensus guidelines for pancreatic imaging. A 2024 white paper from these groups recommended specific pulse sequences, quality control phantoms, and reporting templates for pancreatic MRI in diabetes, representing an important step toward harmonization. Similarly, the World Federation for Ultrasound in Medicine and Biology has convened a panel to establish elastography acquisition protocols, including recommendations for minimum number of valid measurements, acceptable inter-measurement variability, and appropriate reporting units.

Integration Into Clinical Workflow

Even validated techniques face logistic barriers to adoption in routine diabetes care. Adding a 30-minute MRI scan to a routine diabetes visit requires scheduling coordination, insurance preauthorization, and often a separate appointment, which reduces patient adherence. Similarly, biomarker panels that require central laboratory processing with 48–72 hour turnaround times may be impractical for clinics that make immediate therapeutic decisions. Point-of-care devices — handheld elastography probes and lateral-flow assays for miR-375 or unmethylated INS DNA — are in development and could bypass these bottlenecks. Several companies are commercializing compact, cartridge-based systems that can quantitate multiple biomarkers from a fingerstick sample within 15 minutes. A 2024 pilot study in a real-world diabetes clinic demonstrated that a point-of-care miR-375 assay, combined with point-of-care HbA1c testing, changed management decisions in 22% of cases compared with standard care alone.

Adoption will also require updated clinical practice guidelines from bodies like the American Diabetes Association (ADA) and European Association for the Study of Diabetes (EASD). Currently, the ADA's Standards of Medical Care in Diabetes do not include specific recommendations for non-invasive pancreatic function testing, beyond C-peptide measurement. Incorporating these emerging tools into future guidelines will need robust evidence of improved clinical outcomes — not just correlation with established measures — as well as cost-effectiveness data and feasibility assessments. Professional societies can accelerate this process by organizing consensus conferences and commissioning systematic reviews that evaluate the weight of evidence.

Regulatory and Reimbursement Hurdles

Few non-invasive pancreatic function tests have received FDA or EMA approval specifically for diabetes management. Ultrasound elastography is approved for liver fibrosis evaluation but is often used off-label for the pancreas, creating medicolegal uncertainty for clinicians. MRI sequences such as T1 mapping and DWI are available on commercial scanners but lack specific labeling for pancreatic fibrosis quantification. Obtaining dedicated regulatory clearance requires expensive trials that demonstrate safety, analytic validity, clinical validity, and clinical utility — a high bar that has slowed commercialization. Reimbursement is a related challenge: payers may deny coverage for off-label uses or for tests they consider investigational, even when published evidence supports their clinical value. In the US, the Centers for Medicare & Medicaid Services (CMS) has not established a specific reimbursement code for pancreatic elastography, and private insurers vary widely in their coverage policies. Advocacy from professional societies, together with pilot real-world evidence programs that demonstrate improved outcomes and reduced costs, are gradually shifting this landscape. The FDA's Medical Device Development Tools program offers a pathway for qualifying non-invasive biomarkers as drug development tools, which could accelerate their acceptance even before formal labeling expansion.

Future Innovations on the Horizon

Artificial Intelligence Interpretation

Deep learning algorithms trained on thousands of pancreatic MRI scans can automatically segment pancreatic parenchyma, quantify fibrosis scores, and predict beta-cell failure risk with accuracy that rivals expert radiologists. Convolutional neural network architectures, such as U-Net variants, achieve Dice similarity coefficients above 0.90 for pancreatic segmentation — a critical first step for downstream quantification. When combined with radiomics feature extraction — encompassing hundreds of textural, shape, and intensity-based features from segmented images — these models can capture subtle parenchymal changes invisible to the human eye. A 2024 model published in Nature Digital Medicine integrated MRI radiomics with clinical variables to predict progression from impaired fasting glucose to type 2 diabetes within 3 years, achieving an AUC of 0.94 in a held-out test set. As these tools mature, they will reduce radiologist workload, improve consistency across interpretations, and bring expert-level assessment to hospitals and clinics without subspecialty expertise in pancreatic imaging.

AI is also being applied to biomarker panel interpretation. Multivariate machine learning algorithms, including random forests and gradient-boosted trees, can integrate dozens of circulating biomarkers, clinical variables, and imaging parameters into a single risk score that outperforms any individual test. A 2023 study demonstrated that an AI model incorporating miR-375, unmethylated INS DNA, C-peptide, and HbA1c predicted progression of type 1 diabetes with greater accuracy than the Diabetes Prevention Trial–Type 1 Risk Score, a widely used logistic regression model. The flexibility of these algorithms allows them to accommodate missing data — for example, if a particular biomarker assay fails or is unavailable — and still generate a valid risk prediction, enhancing their clinical robustness.

Wearable and Implantable Sensors

Continuous glucose monitors (CGMs) already provide indirect feedback on beta-cell function through metrics such as glucose variability, time in range, and, more recently, glucose disappearance rate after mixed meals, which correlates with insulin secretion capacity. However, researchers are now exploring sweat-based biosensors for direct detection of C-peptide and insulin. Wearable patches that collect sweat via iontophoresis or passive diffusion, combined with enzymatic or immunoassay-based detection, have demonstrated preliminary feasibility in proof-of-concept studies. A 2024 report in Biosensors and Bioelectronics described a flexible wristband that could measure sweat C-peptide within a clinically relevant range (1–10 ng/mL) with accuracy within 15% of serum C-peptide measured by ELISA.

Microneedle patches that sample interstitial fluid represent a more direct approach. These arrays of microscopic needles — typically less than 1 millimeter in length — penetrate the stratum corneum without stimulating pain fibers, enabling painless access to dermal interstitial fluid. Microneedle-based biosensors have been developed for continuous monitoring of glucose, lactate, and pH, and researchers are now extending them to beta-cell stress proteins and microRNAs. A 2023 study demonstrated that a microneedle patch functionalized with anti-miR-375 antibodies could detect changes in miR-375 levels in interstitial fluid that mirrored serum kinetics during induced beta-cell apoptosis in a rodent model. While still preclinical, these technologies promise ambulatory, passive monitoring of pancreatic function during daily life, providing high-density longitudinal data that could revolutionize our understanding of beta-cell dynamics in natural history and in response to interventions.

Combined Multi-Omics Profiles

Integrating imaging findings with circulating metabolites, lipids, proteins, and RNA signatures offers a systems-level view of pancreatic health. A "pancreatomics" approach — analogous to the "hepatomics" framework that has emerged for liver disease — could eventually generate a single composite risk score that replaces multiple separate tests. Early proof-of-concept studies have shown the power of this approach. A 2024 study in Nature Medicine integrated MRI elastography, plasma miR-375 levels, and stool metagenomic profiles to predict progression from prediabetes to type 2 diabetes in a cohort of 800 participants. The combined model achieved 92% accuracy in a held-out validation set, compared with 78% for clinical variables alone (including age, BMI, HbA1c, and fasting glucose). Importantly, the model identified a subset of participants with low clinical risk but high molecular and imaging risk who progressed rapidly — individuals who would have been missed by conventional screening.

Such integrated approaches require sophisticated data harmonization, cross-modality normalization, and machine learning architectures capable of handling heterogeneous data types. Graph neural networks and multi-modal transformers are emerging as effective frameworks for this task, allowing the model to learn relationships between imaging features, molecular biomarkers, and clinical outcomes. The success of these models will depend on the availability of large, well-annotated cohorts that span the full spectrum of diabetes — from autoantibody-positive preclinical stages to established disease with complications. International collaborative efforts, such as the Innovative Medicines Initiative's Diabetes Research on Patient Stratification and the National Institutes of Health's Human Pancreas Analysis Program, are generating exactly these types of multi-modal datasets, accelerating progress toward a unified, non-invasive assessment of pancreatic function.

Conclusion

Non-invasive techniques for evaluating pancreatic function in diabetes are advancing rapidly, driven by technological innovation in imaging, biomarker discovery, and computational analysis. MRI elastography, ultrasound shear-wave imaging, salivary and blood microRNA panels, novel contrast approaches, and emerging multi-omics integration each offer safer, repeatable, and often more accessible alternatives to the invasive tests that have historically defined the standard of care. While challenges in validation across diverse populations, standardization of protocols, clinical workflow integration, and regulatory/reimbursement pathways remain active areas of work, the trajectory is clear: the future of pancreatic function assessment will be non-invasive, multimodal, and increasingly driven by artificial intelligence and point-of-care devices.

These innovations promise to transform diabetes management by enabling earlier detection of beta-cell dysfunction, more precise monitoring of disease progression, and personalized therapy selection — all while eliminating the discomfort, risk, and logistical burden of older methods. As these tools mature from research settings into clinical practice, they will not only improve outcomes for individuals living with diabetes but also facilitate the development of novel therapies by providing sensitive, dynamic endpoints for clinical trials. The next decade holds the promise of a comprehensive, non-invasive pancreatic health assessment that is as routine and accessible as a blood pressure measurement or a lipid panel.

For further reading on specific techniques, see the recent meta-analysis on ultrasound elastography in diabetes in PubMed, the cost-effectiveness analysis in Diabetes Care, the review of salivary biomarkers in Journal of Diabetes Research, and the multi-omics study in Nature Medicine.