REGISTER
ADA 2026: 7 Studies From Prof.Jianping Weng's Team on Diabetes Heterogeneity
At ADA 2026, Prof. Jianping Weng and Prof. Xueying Zheng 's team from the First Affiliated Hospital of USTC presented 7 studies spanning T1DM heterogeneity classification, multi-omics profiling of T2DM complications, metabolic-inflammatory interaction mechanisms, and digital health interventions — advancing a precision medicine paradigm for complex metabolic diseases.

During the 2026 American Diabetes Association Scientific Sessions (ADA 2026), Prof. Weng Jianping and Prof. Zheng Xueying's team from the First Affiliated Hospital of the University of Science and Technology of China (USTC) (Anhui Provincial Hospital) presented 7 research advances covering T1DM heterogeneity subtyping, multi-omics profiling of T2DM complications, metabolic-inflammatory interaction mechanisms, and novel digital health intervention approaches. The following are the main research abstracts.

Dynamic Evolution of Social Support Behaviors in the T1DM Online Community Based on Hidden Markov Models

Abstract No.: 1630-P

Researchers: Zhu Yujia, Zhao Chenxiao, Zeng Xifeng, Zhang Yanxiang, Zheng Xueying

Online health communities (OHCs) play an important role in providing social support for patients with chronic diseases, but the longitudinal dynamics of support behaviors have not been well characterized. This study aimed to analyze longitudinal patterns and influencing factors underlying the transition from "seeking support" to "providing support" among users of China's largest T1DM online community.

A longitudinal analysis was conducted on 378,304 posts from 7,803 users between 2019 and 2024. The BERT-Base-Chinese model was used to classify posts into informational support seeking/providing and emotional support seeking/providing behaviors; Chinese Clinical BERT was used to identify emotional intensity. A hidden Markov model (HMM) was used to model behavioral state transitions (baseline state, support-seeking state, support-providing state), incorporating disease duration and community engagement as covariates, with a 6-month time window. The Viterbi algorithm was used to reconstruct individual behavioral trajectories, and dynamic time warping (DTW)-based clustering was applied to analyze trajectory patterns.

Results showed that 46.2% of users experienced at least one behavioral state transition. Transitions from support-seeking to support-providing (28.4%) were more common than the reverse (9.7%). Longer disease duration (β = 0.42, P < 0.01) and longer community engagement (β = 0.31, P < 0.05) were positively associated with the likelihood of becoming a support provider. Higher emotional intensity was also positively associated with this transition (β = 0.27, P < 0.05). Three typical behavioral trajectory types were identified: emotionally driven (32%), rationally accumulated (45%), and fluctuating (23%).

These findings indicate a clear longitudinal shift from support-seeking to support-providing among T1DM online community users, closely associated with accumulated disease experience and sustained community engagement.

C-Reactive Protein–Triglyceride–Glucose Index and Risk of New-Onset Chronic Disease and Mortality: Evidence from Multi-Country Cohorts

Abstract No.: 2229-P

Researchers: Li Xingyu¹, Zhang Hongqiang¹, Zhang Junhao¹, Shi Jie¹, Wang Suijian², Zhao Minzhe¹, He Meiwen², Yuan Zhou¹, Li Chenghua³, Zheng Xueying¹, Ding Yu¹, Yue Tong¹

¹Department of Endocrinology and Metabolism, First Affiliated Hospital of USTC (Anhui Provincial Hospital); ²Department of Endocrinology, First Affiliated Hospital of Anhui Medical University; ³Pan-Vascular Disease Management Center, First Affiliated Hospital of USTC (Anhui Provincial Hospital)

Chronic inflammation and insulin resistance contribute to the development of multiple chronic diseases. The C-reactive protein–triglyceride–glucose index (CTI) integrates inflammation, lipid, and glucose metabolism information to reflect an individual's metabolic-inflammatory burden. This study investigated the association between CTI and the risk of chronic disease onset and mortality across multi-country cohorts and evaluated its predictive value.

CHARLS (n = 5,366) and NHANES (n = 2,921) were used as the primary cohorts, with validation in the UK Biobank and hospital cohorts. Cox regression was used to analyze the associations of CTI and cumulative CTI with new-onset chronic disease and mortality risk. K-means clustering, restricted cubic splines, and area under the curve (AUC) were used to evaluate CTI change patterns, non-linear associations, and predictive performance.

Results showed that elevated CTI was closely associated with increased risk of chronic disease onset and mortality. In the CHARLS cohort, higher cumulative CTI was associated with greater risk of new-onset hypertension (HR: 1.18, 95% CI: 1.11–1.26), diabetes (HR: 1.60, 95% CI: 1.47–1.74), dyslipidemia (HR: 1.80, 95% CI: 1.69–1.92), stroke (HR: 1.15, 95% CI: 1.03–1.29), and asthma (HR: 1.25, 95% CI: 1.08–1.45). In the NHANES cohort, higher CTI levels were associated with increased risk of all-cause mortality (HR: 1.19, 95% CI: 1.14–1.25) and mortality related to cancer, heart disease, pulmonary disease, and diabetes. Mortality risk was markedly elevated when CTI exceeded 9.0. CTI demonstrated good predictive ability for dyslipidemia and mortality, with AUC values of 0.74 and 0.85, respectively. Results were validated in two independent cohorts.

In summary, elevated CTI is associated with increased risk of multiple new-onset chronic diseases and mortality. As a composite index integrating inflammation and metabolic information, CTI demonstrates good risk identification and predictive value.

Multi-omics Atlas of Multi-System Complications in Type 2 Diabetes

Abstract No.: 2283-P

Researchers: Zhang Hongqiang¹, Yue Tong¹, Wang Suijian², Shi Jie¹, Geng Mengya¹, Zhang Haoqiang¹, Pan Lanxin¹, Gu Zhaohe¹, Sun Lijie¹, Zhao Minzhe¹, Zheng Xueying¹, Weng Jianping³

¹Department of Endocrinology and Metabolism, First Affiliated Hospital of USTC (Anhui Provincial Hospital); ²Department of Endocrinology, First Affiliated Hospital of Anhui Medical University; ³Chair Professor, USTC; current President, Anhui Medical University

T2DM can cause multi-system complications, yet an integrated multi-omics research framework for cross-system, multi-outcome analysis has been lacking. This study aimed to systematically construct a proteomics and metabolomics atlas of major T2DM-related outcomes and identify molecular prediction panels with both predictive performance and clinical feasibility.

Based on a prospective T2DM cohort, separate proteomics, metabolomics, and multi-omics cohorts were established. Through cross-sectional and longitudinal analyses, the associations between plasma proteins and metabolites and 19 T2DM-related outcomes were systematically evaluated. Machine learning-based molecular feature selection strategies were further used to build risk prediction models, which were compared with conventional clinical risk models.

Results identified a series of molecular signals showing consistent positive or negative associations across multiple T2DM-related outcomes, suggesting shared biological pathways. Different outcomes also displayed specific and heterogeneous molecular signatures. Predictive analyses showed that proteomics-based models significantly outperformed clinical models; combined models achieved the best overall predictive performance, with consistent improvement on reclassification metrics. Metabolomics features provided only limited incremental predictive benefit. A simplified prediction panel of approximately one hundred proteins was identified that maintained robust predictive performance.

This large-scale multi-omics study systematically constructed a molecular atlas of T2DM complications, offering new insights into disease pathogenesis and potential therapeutic targets, and proposing a risk prediction and precision intervention framework with both clinical feasibility and practical utility.

DDRTree Dimensionality Reduction Reveals Age of Onset as a Key Phenotype of T1DM Heterogeneity and Supports the Existence of Age-of-Onset-Related Endotypes

Abstract No.: 2371-P

Researchers: Chang Shiru, Tan Huiling, Yue Tong, Ding Yu, Gu Zhaohe, Shi Yu, Pan Lanxin, Guo Chuang, Weng Jianping, Zheng Xueying

T1DM is a highly heterogeneous disease, and the key phenotypes that characterize this heterogeneity remain unclear. This study employed a tree-based dimensionality reduction approach (DDRTree) to identify key clinical phenotypes underlying T1DM heterogeneity and to validate the existence of age-of-onset-related endotypes.

The study enrolled 894 T1DM patients and performed DDRTree dimensionality reduction based on 13 phenotypes including age of onset, BMI, blood pressure, ALT, HbA1c, creatinine, triglycerides, HDL cholesterol, total cholesterol, and 3 islet autoantibodies. Age-of-onset endotype cutoffs (< 7 years, 7–13 years, ≥ 13 years) were validated based on DDRTree coordinates, and the clinical, immunological, and genetic characteristics of patients in each endotype were systematically characterized. Cardiovascular outcome risks across endotypes were further explored.

Results showed that age of onset was the key phenotype characterizing T1DM heterogeneity, explaining 64.6% of variance in the dimensionality reduction results. Based on DDRTree coordinates, significant differences were observed between age-of-onset groups, supporting < 7 years, 7–13 years, and ≥ 13 years as endotype cutoff values. Significant differences were observed across endotypes in autoantibody positivity rate (P < 0.001), genetic risk (P < 0.001), and metabolic features. Cardiovascular outcome analysis showed that compared with matched healthy controls, both the < 7 years and ≥ 13 years onset groups had elevated cardiovascular risk, with the < 7 years onset group showing more pronounced excess risk; cardiovascular risk ratios were 4.40 (95% CI: 2.78–6.95) and 2.67 (95% CI: 2.31–3.09) for the two groups, respectively.

The study demonstrates that age of onset is a key phenotype characterizing T1DM heterogeneity and supports the existence of age-of-onset-related endotypes. Endotypes differ in immune features, genetic risk, and long-term cardiovascular risk.

Deep Learning Identification of Multi-omics Features and Associated Mechanisms in DKA-Onset T1DM

Abstract No.: 2389-P

Researchers: Tan Huiling, Yue Tong, Ding Yu, Gu Zhaohe, Shi Yu, Pan Lanxin, Chang Shiru, Guo Chuang, Weng Jianping, Zheng Xueying

T1DM presenting with diabetic ketoacidosis (DKA) at onset versus those without show significant pathophysiological heterogeneity; the role of environmental factors such as gut microbiota in this heterogeneity has not been established. Conventional approaches have difficulty systematically decoding complex inter-omics associations.

The study enrolled 69 newly diagnosed T1DM patients divided into DKA and non-DKA groups according to DKA status at onset. A deep learning multi-omics variational autoencoder (MOVE) was used to systematically evaluate group differences in gut microbiome, serum metabolome, lipidome, and proteome characteristics, and perturbation analyses were performed to identify key microbiota and their associations with metabolic/protein features.

The MOVE model demonstrated excellent performance in processing T1DM multi-omics data (loss = 0.9403). The two groups displayed distinct multi-omics signatures: the DKA group was characterized by elevated Bacteroides clarus, 2-arachidonoylglycerol ether, and phosphatidylcholine [PC (36:2)]; the non-DKA group was characterized by elevated Burkholderiaceae, D-(+)-proline, lysophosphatidylcholine [LPC (16:0)], and CCL28 protein levels. Integrative analysis confirmed that gut microbiota plays a key role in DKA onset. Perturbation analysis suggested that a specific species of Lachnospira (Lachnospira sp000437735) may improve DKA status at T1DM onset by promoting the metabolism of benzyl alcohol to benzaldehyde.

The study demonstrates that the application of deep learning models provides a novel molecular perspective for decoding the heterogeneity of T1DM onset and offers potential therapeutic targets for individualized prevention strategies.

Multivariate GWAS Identifies Pleiotropic Loci Shared Between Obesity and MASLD

Abstract No.: 2615-P

Researchers: Geng Mengya¹², Wang Suijian⁴, Zhang Hongqiang¹, Sun Lijie¹, Weng Jianping³⁴, Zheng Xueying¹

¹Department of Endocrinology and Metabolism, First Affiliated Hospital of USTC (Anhui Provincial Hospital); ²Clinical Medical College, Wannan Medical College; ³School of Life Sciences and Medicine, USTC; ⁴Anhui Medical University

Obesity and metabolic dysfunction-associated steatotic liver disease (MASLD) are highly comorbid conditions, but their shared genetic architecture and tissue-specific regulatory mechanisms are not fully understood. This study used an integrative approach combining multiple genetic epidemiological and functional genomic methods to systematically characterize the genetic overlap between obesity-related phenotypes and MASLD, identifying potential pleiotropic loci and candidate regulatory genes.

The study utilized publicly available European population GWAS summary data, incorporating 4 obesity-related phenotypes (BMI, BMI-adjusted waist circumference, BMI-adjusted waist-hip ratio, BMI-adjusted hip circumference) and NAFLD/MASLD-related phenotypes. Linkage disequilibrium score regression (LDSC) was used to evaluate genome-wide genetic correlation; conditional false discovery rate (condFDR) and conjunctional FDR (conjFDR) were applied to identify single-phenotype and shared genetic loci; and multi-trait analysis of GWAS (MTAG) was used to enhance detection power. eQTL data from liver, subcutaneous adipose, and visceral adipose tissue from GTEx v8 were combined with summary-data-based Mendelian randomization (SMR) to explore potential causal links between gene expression and obesity/MASLD phenotypes.

Genetic correlation analysis showed a significant positive correlation between BMI and MASLD (rg = 0.687, P = 1.30 × 10⁻⁶); BMI-adjusted waist-hip ratio was also positively correlated with MASLD (rg = 0.522, P = 3.00 × 10⁻⁴), while BMI-adjusted hip circumference was negatively correlated with MASLD (rg = −0.273, P = 9.90 × 10⁻³). condFDR analysis identified 31 independent MASLD-associated loci, 15 of which were novel and mapped to 13 genes; 184 obesity-associated independent loci were also identified. conjFDR revealed 23 obesity-MASLD shared loci. MTAG analysis identified 174 MASLD-associated risk loci and 482 obesity-associated risk loci, respectively. Cross-method integration showed that ADCY3, FTO, TRIM66, COBLL1, L3MBTL3, VPS53, CDK6, ANAPC13, and TBX15 received support from multiple analytical approaches and may represent important pleiotropic genetic signals linking obesity and MASLD. SMR analysis suggested that some genes exert regulatory effects of differing direction or magnitude in liver, subcutaneous adipose, and visceral adipose tissue, reflecting possible tissue-specific molecular regulatory mechanisms in obesity-MASLD comorbidity.

This study confirms significant genetic overlap between obesity and MASLD, identifies a set of potential pleiotropic loci and tissue-specific regulatory genes, and provides new leads for understanding the genetic mechanisms of obesity-related MASLD, optimizing risk stratification in high-risk populations, and exploring intervention targets for metabolic diseases.

Exosomal miR-223-3p Targeting NLRP3-Associated Neuroinflammation as a Potential Biomarker for Cognitive Dysfunction in T2DM

Abstract No.: 1117-OR

Researcher: Zhang Haoqiang

Department of Endocrinology and Metabolism, First Affiliated Hospital of USTC

Neuroinflammation plays an important role in diabetic cognitive dysfunction, with the NLRP3 inflammasome as a key mediator. This study aimed to elucidate the regulatory role of miR-223-3p in NLRP3-associated neuroinflammation and explore its potential as a biomarker for mild cognitive impairment (MCI) in T2DM patients.

A diabetic mouse model was established by high-fat diet feeding for 24 weeks, and cognitive impairment was confirmed by behavioral assessment. MicroRNA sequencing of hippocampal tissue was performed to validate miR-223-3p expression changes. NLRP3-mediated neuroinflammation was examined in vivo; mechanistic and pathway studies were conducted in vitro using BV2 cells. The effects of miR-223-3p knockdown and overexpression on NLRP3 were evaluated. Clinically, plasma exosomal miR-223-3p levels were measured in T2DM patients with and without MCI, and their correlations with neuropsychological test scores were analyzed.

Results showed that diabetic mice exhibited cognitive decline, reduced hippocampal miR-223-3p expression, microglial activation, and elevated NLRP3 and pro-inflammatory cytokine levels. miR-223-3p knockdown induced neuroinflammation, while overexpression attenuated it. Luciferase reporter assays confirmed direct regulation of NLRP3 by miR-223-3p. Clinically, exosomal miR-223-3p levels were significantly lower in T2DM patients with MCI compared with cognitively normal patients, and correlated with overall cognitive function and immediate memory performance. ROC analysis showed that at a threshold of 0.7565, miR-223-3p had a specificity of 71.4% and sensitivity of 68.7% for diagnosing MCI.

In conclusion, miR-223-3p exerts neuroprotective effects by suppressing NLRP3-mediated neuroinflammation. Its downregulation contributes to the development of diabetic cognitive impairment, and circulating exosomal miR-223-3p may serve as a potential biomarker for identifying MCI in T2DM patients.

Commentary by Prof. Weng Jianping

At ADA 2026, our team presented 7 studies that, while spanning multiple areas — T1DM heterogeneity, multi-omics profiling of T2DM complications, metabolic-inflammatory interactions, and digital health intervention — are unified by a single thread: diabetes is accelerating its transition from empirically managing a "single disease" to a new paradigm of precision medicine grounded in multidimensional data integration.

What strikes me most is that these studies collectively converge on a core proposition — heterogeneity. Whether it is the age-of-onset endotypes revealed by DDRTree, the multi-omics signatures of DKA onset decoded by the MOVE model, or the risk stratification of chronic disease comorbidity by the CTI index, all are answering the same question: behind the same diagnostic label of "diabetes" lie fundamentally different molecular mechanisms, disease trajectories, and treatment responses. In the past, we managed patients using blood glucose as a single metric with a one-size-fits-all approach; in the future, we must build a multidimensional stratification system encompassing genomics, proteomics, metabolomics, microbiomics, and even digital behavioral phenotypes.

Also worth deep reflection is the pervasive integration of intelligent digital technologies. The dynamic capture of online community support behaviors via hidden Markov models is not merely a digital phenotyping study — it signals that chronic disease management is breaking through hospital walls and entering the continuum of "community–home–digital space." When multi-omics atlases meet deep learning, and clinical cohort data meet real-world behavioral data, we are effectively constructing the prototype of a "virtual artificial pancreas" and a "metabolic digital twin" — this is precisely the clinical translation direction our team aims to advance next.

From a broader perspective, diabetes research is undergoing three transitions: from single-disease to pan-vascular metabolic comorbidity networks; from endpoint-driven treatment to full-lifecycle risk prediction; and from Western population evidence to China-specific solutions. The multi-country cohort validation of CTI, the obesity-MASLD genetic overlap analysis, and the early biomarker exploration for cognitive impairment presented at this ADA all provide foundational evidence for these transitions.

I have always believed that Chinese diabetes research cannot merely follow international guidelines. When our T1DM large models, artificial pancreas algorithms, and multi-omics prediction panels begin generating "Chinese parameters," we are in fact participating in defining the global standard for the next generation of diabetes management. These 7 ADA findings are both a milestone summary of our team's prior work and stepping stones toward the era of "computational precision therapy."

Team Research Directions and Achievements

Prof. Weng Jianping's team has long focused on precision diabetes diagnosis and clinical translation, building a systematic research framework across three dimensions:

1. Exploring disease mechanisms with intelligent digital tools: The team established China's first intelligent management and big data center for T1DM, built the world's first disease-specific large model for T1DM, and developed and validated a novel artificial pancreas algorithm and a physics-informed personalized linear model control system (patent granted: ZL202510565116.8). By integrating deep learning, hidden Markov models, multi-omics variational autoencoders (MOVE), and other AI technologies, the team has established a digital research framework spanning molecular mechanism analysis to clinical risk prediction, driving intelligent and precision transformation of diabetes prevention and treatment.

2. Scientific research on diabetic complications and comorbidities: Based on large-scale prospective cohorts, the team systematically conducted multi-omics profiling of multi-system complications in T2DM, integrating proteomics, metabolomics, and lipidomics data to construct molecular prediction panels for 19 related outcomes. DDRTree dimensionality reduction revealed age of onset as the key phenotype of T1DM heterogeneity, validating differences in cardiovascular outcome risk across endotypes. Multi-country cohort studies of CTI and chronic disease/mortality risk were conducted; genetic overlap between obesity and MASLD and tissue-specific regulatory mechanisms were investigated; and the molecular mechanism of exosomal miR-223-3p involvement in diabetic cognitive dysfunction via the NLRP3 pathway was elucidated.

3. Serving clinical practice: The team's research findings have been cited in 35 international guidelines/consensus documents and authoritative textbooks worldwide. The team has led the development of multiple clinical guidelines and group standards. Based on the "beta-cell rest" hypothesis, a new paradigm for rational insulin use was established; a 24-year nationwide multicenter cohort study demonstrated that early insulin therapy reduces the risk of stroke by 31% and the risk of heart failure hospitalization by 28%. The team created a pan-vascular disease management center to advance integrated metabolic-vascular management; constructed an online health community behavioral evolution model to provide evidence-based support for digital health interventions in chronic diseases; and developed a clinically feasible risk prediction and precision intervention framework, directly serving clinical decision-making for high-risk population screening, risk stratification, and early intervention.

0
0
Loading