Olink proteomics identified 84 differentially expressed proteins in T2DM patients, with a three-protein model (VSIG2, S100A11, S100A5) achieving favorable discriminative performance (AUC=0.874) for age stratification and biological senescence evaluation in T2DM, superior to conventional clinical indicators.
Key Findings
Results
84 differentially expressed proteins were identified between T2DM patients and healthy controls, enriched in cytokine-cytokine receptor interaction, lipid metabolism, and atherosclerosis pathways.
Study enrolled 21 healthy controls and 66 T2DM patients
Plasma protein profiles were detected via Olink targeted proteomics
All 84 DEPs had P < 0.05
GO and KEGG functional enrichment analyses were applied to characterize the DEPs
Key enriched pathways included cytokine-cytokine receptor interaction, lipid metabolism, and atherosclerosis
Results
Machine learning identified six core proteins with high discriminatory value for T2DM: MERTK, BOC, TNFRSF10A, CCL3, AGRP, and PD-L2.
Random forest and LASSO regression analyses were applied to screen core T2DM-related protein molecules
Six proteins — MERTK, BOC, TNFRSF10A, CCL3, AGRP, and PD-L2 — were identified as core discriminatory biomarkers
These proteins were selected from the broader set of 84 DEPs
The combination of two complementary machine learning methods was used to improve selection robustness
Results
Nine age-dependent plasma proteins were independently identified within the T2DM cohort, with VSIG2, LPL, S100A11, and CST5 showing the most robust correlations with age.
T2DM patients were stratified using a 55-year cutoff, described as 'a well-recognized critical threshold for metabolic senescence'
Nine proteins were identified as age-dependent after adjustment for sex, BMI, HbA1C, and hypoglycemic medication use (FDR < 0.05)
VSIG2, LPL, S100A11, and CST5 exhibited the most robust correlations with age (FDR < 0.01)
Identification was performed after adjustment for potential confounders to isolate age-independent senescence-related proteins
Results
A three-protein model comprising VSIG2, S100A11, and S100A5 achieved an AUC of 0.874 for discriminating age groups within the T2DM cohort.
The three-protein model AUC of 0.874 was superior to BMI (AUC = 0.572) and HbA1c (AUC = 0.593)
Bootstrap validation used a 1000-time resampling procedure for internal validation
Bias-corrected AUC was 0.900 and the correction optimism was -0.024
Overfitting degree was -2.65%, indicating reliable model stability
The model was described as providing 'favorable discriminative performance'
Results
Candidate biomarker proteins were functionally associated with immune homeostasis, lipid remodeling, inflammatory responses, and calcium signaling.
Functional analyses were conducted on the candidate proteins identified through machine learning and age-stratification analyses
Key biological processes implicated included immune homeostasis, lipid remodeling, inflammatory responses, and calcium signaling
LPL (lipoprotein lipase) is a known lipid metabolism enzyme, consistent with lipid remodeling involvement
S100A11 and S100A5 are calcium-binding proteins, consistent with calcium signaling involvement
Results
Routine clinical indicators HbA1c and BMI showed poor discriminatory performance for age-based biological senescence stratification in T2DM patients.
BMI had an AUC of 0.572 for age group discrimination within the T2DM cohort
HbA1c had an AUC of 0.593 for age group discrimination within the T2DM cohort
Both values are substantially lower than the three-protein model AUC of 0.874
The authors concluded that the proposed model 'compensates for the limitations of routine clinical indices'
What This Means
This research suggests that standard blood tests used to monitor diabetes — such as HbA1c (a measure of long-term blood sugar control) and BMI — are poor tools for distinguishing biological aging burden in people with type 2 diabetes. Using a high-throughput protein measurement technology called Olink proteomics, the researchers measured hundreds of proteins in the blood of 66 diabetes patients and 21 healthy individuals, ultimately identifying 84 proteins that differed significantly between the two groups. Using machine learning techniques, they narrowed this down to six proteins most useful for identifying diabetes itself, and separately identified nine proteins whose levels changed with age in diabetic patients.
The study's most practical finding is a three-protein combination — VSIG2, S100A11, and S100A5 — that could distinguish younger from older diabetic patients (using age 55 as a biological aging threshold) far better than conventional clinical markers. This three-protein model had an accuracy score (AUC) of 0.874, compared to just 0.572 for BMI and 0.593 for HbA1c. The model remained stable when tested through a rigorous 1000-repetition statistical validation process, with an overfitting degree of only -2.65%.
This research suggests that measuring specific blood proteins may provide a more precise way to assess how much biological aging has occurred in a person with diabetes, which could eventually help doctors tailor treatments more appropriately to an individual patient's actual biological age rather than their calendar age. The proteins identified are linked to immune function, fat metabolism, inflammation, and calcium signaling — processes known to be disrupted in both aging and diabetes. The study was conducted in a relatively small Chinese patient cohort, so larger and more diverse studies would be needed before these biomarkers could be used clinically.
Yang Y, Liu S, Xie C, Zhu D. (2026). Olink proteomics identifies serum protein biomarkers and an age-stratified diagnostic model for diabetes in Chinese patients.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1900153