Urinary extracellular vesicle miRNA signature reflects pancreatic islet stress in type 2 diabetes

ABSTRACT

Introduction

Type 2 diabetes (T2D) is a progressive metabolic disorder characterized by insulin resistance and progressive β‐cell dysfunction. Early detection remains critical to prevent long‐term complications. Urinary extracellular vesicle (ECV) microRNAs (miRNAs) have emerged as stable, non‐invasive biomarkers with the potential to reflect systemic molecular alterations associated with metabolic disease.

Methods

We analyzed previously generated urinary ECV miRNA sequencing data from a well‐characterized cohort of 68 adults (40 T2D and 28 healthy controls). Differentially expressed miRNAs were identified and evaluated for diagnostic performance using receiver operating characteristic (ROC) analysis and supervised machine learning models with 10‐fold cross‐validation. Independent external validation was performed to assess generalizability. Cross‐tissue validation was conducted using publicly available datasets from pancreatic islets, blood, liver, and adipose tissue. Predicted target genes were examined across tissues, and miRNA–mRNA interaction networks with pathway enrichment analyses were performed to explore functional relevance.

Results

Forty‐six miRNAs were significantly dysregulated in urinary ECVs from T2D patients compared with controls. Network bottleneck centrality analysis prioritized five key miRNAs (miR‐320a, miR‐16‐5p, miR‐125b‐5p, miR‐26a‐5p, and miR‐30c‐5p). Individual miRNAs demonstrated moderate discriminatory capacity (AUC 0.73–0.81), while the combined panel improved performance (internal AUC = 0.87; external AUC = 0.86). Dysregulated urinary miRNA patterns partially mirrored expression changes in pancreatic islets and other metabolic tissues. Target gene analysis revealed tissue‐specific alterations in key metabolic regulators, including PTEN, IGF1R, HMGA1, VEGFA, MCL1, CCND2, BTG2, and SMAD4.

Conclusion

Urinary ECV miRNAs reflect molecular alterations associated with T2D and represent promising complementary, non‐invasive biomarkers with mechanistic relevance to disease progression.

Urinary ECV miRNAs show a distinct dysregulation signature in T2D with strong diagnostic performance. This signature mirrors molecular changes across pancreatic islets, blood, liver, and adipose tissue, with the strongest concordance in islets, linking it to β‐cell stress. Overall, they represent a non‐invasive, mechanistically informative biomarker of systemic disease remodeling.

INTRODUCTION

Type 2 diabetes (T2D), characterized by chronic hyperglycemia due to insulin resistance and β‐cell dysfunction, leads to serious complications such as retinopathy, nephropathy, and neuropathy if left uncontrolled ¹ . Globally, the prevalence of T2D is rising sharply, with the Middle East and North Africa (MENA) region experiencing an especially high burden ² . According to the International Diabetes Federation, 73 million adults in the MENA region had diabetes in 2021, a number predicted to reach 95 million by 2030 ³ . Kuwait, with an adult diabetes prevalence of 25.5%, the highest in the world, highlights the severity of this public health challenge ⁴ . This epidemic contributes to substantial morbidity, places a significant burden on healthcare systems, and leads to premature mortality.

There is a pressing need for non‐invasive biomarkers to facilitate early diagnosis and timely intervention ⁵ . Although traditional blood‐based markers such as HbA1c and fasting glucose are widely used, their diagnostic accuracy can be influenced by ethnic variability, red blood cell turnover, and hemoglobin glycation rates. This is especially relevant in populations like Kuwait, where T2D may develop at lower BMI or at younger ages ⁴ , ⁶ . Moreover, while major diagnostic frameworks use similar thresholds, they differ in how they incorporate ethnic‐specific risk factors ⁷ , potentially leaving some individuals in a diagnostic “gray zone” where early disease may go undetected ⁸ .

In response to these limitations, urine‐based biomarkers have garnered attention for their potential to improve the detection of diabetes. Among these, albuminuria is widely used as a marker for diabetic nephropathy (DN), yet it typically reflects late‐stage kidney damage ⁹ . Proteomic studies have identified inflammatory markers in the urine of patients with early‐stage diabetes, suggesting that immune dysregulation occurs well before overt proteinuria ¹⁰ . Urinary ECVs are enriched in stable, tissue‐derived miRNAs, making them a biologically informative and relatively protected source for biomarker discovery. Moreover, ECVs offer a promising alternative source of biomarkers. These vesicles encapsulate diverse molecular cargos, including microRNAs (miRNAs) ¹¹ , which not only mirror the underlying pathophysiological changes in T2D but are also being investigated as indicators of late complications, such as DN ¹² .

Although miRNA biomarkers have been studied in the context of T2D complications, relatively few investigations have applied next‐generation sequencing (NGS) to profile urinary ECV‐associated miRNAs for early‐stage T2D diagnosis. To address this gap and the growing interest in miRNA‐based diagnostics, this study compares urinary ECV‐associated miRNA profiles between T2D patients without complications and healthy controls in a Kuwaiti cohort. Candidate miRNAs are further evaluated through cross‐validation using publicly available datasets from other populations and tissues relevant to T2D, such as pancreatic islets and blood. This integrative approach supports the use of urinary miRNAs as complementary, non‐invasive biomarkers that may enhance diagnostic precision, particularly when traditional markers are inconclusive, contributing to earlier intervention and better disease management.

MATERIAL AND METHODS

Study cohorts

The data analyzed in this study were previously generated from a well‐characterized cohort of 68 adults, including 40 individuals with type 2 diabetes (T2D) and 28 metabolically healthy controls ¹² . Eligible individuals were adults aged ≥18 years who were able to provide written informed consent. T2D was defined based on established clinical criteria and physician diagnosis. Control participants had no prior diagnosis of diabetes and normal fasting glucose levels. Exclusion criteria included pregnancy, active infection, chronic inflammatory or autoimmune diseases, malignancy, and use of medications known to significantly affect glucose metabolism or inflammatory status.

As described previously, the T2D cohort (n = 40) comprised older individuals. The mean age was 62.4 ± 9.1 years and BMI averaged 34.1 ± 8.1 kg/m². Glycemic parameters were elevated, with fasting glucose of 9.4 ± 3.3 mmol/L and HbA1c of 7.2 ± 1.5%. Renal function was largely preserved, with an albumin‐to‐creatinine ratio (ACR) of 14 ± 7.7 and an eGFR of 81.1 ± 21.2 mL/min/1.73 m². Serum creatinine (68.2 ± 27.1 μmol/L) and BUN (5.7 ± 2.34 mmol/L) were within expected clinical ranges. Lipid parameters showed LDL cholesterol of 1.8 ± 1.1 mmol/L and HDL cholesterol of 1.3 ± 0.4 mmol/L, consistent with a metabolically managed diabetic population. Detailed clinical, biochemical, and demographic characteristics of the study population have been reported previously ¹² . The external validation cohort consisted of 16 T2D patients with age matched to study cohort. Patients were recruited at the Dasman Diabetes Institute in Kuwait. This study was reviewed and approved by the Ethical Review Committee at Dasman Diabetes Institute (Reference: RA/121/2019).

Urine collection, ECVs isolation, RNA extraction, and small RNA sequencing

Urine collection, ECVs isolation, total RNA extraction, small RNA library preparation, and next‐generation sequencing were performed previously and described in detail ¹² . Briefly, urine samples were collected using preservation tubes (Norgen Biotek Corp., Cat. No. 18111), followed by sequential centrifugation and concentration steps to isolate urinary ECVs. ECVs characterization was performed by transmission electron microscopy (TEM), scanning electron microscopy (SEM), and atomic force microscopy (AFM), confirming a size distribution predominantly between 40 and 150 nm. Molecular validation was conducted by flow cytometry, demonstrating positive expression of canonical ECV markers CD63 and TSG101 and negative expression of the endoplasmic reticulum marker GRP94, as described earlier ¹³ . Total RNA was extracted using the Urine EV Purification and RNA Isolation Maxi Kit (Norgen, Canada; Cat. No. 58800). Small RNA libraries were prepared using the QIAseq miRNA Library Kit (Qiagen, Cat. No. 331502), incorporating unique molecular indices (UMIs), and sequenced on the Illumina MiSeq platform (150‐cycle v3 kit). Raw FASTQ files were processed using the GeneGlobe web‐based analysis platform (Qiagen), and normalization was performed using the Trimmed Mean of M‐values (TMM) method implemented in the edgeR package ¹⁴ . Differential expression analysis identified significantly dysregulated miRNAs using a threshold of |log2 fold change| > 2 and P‐value <0.05. Significantly dysregulated miRNAs in T2D patients identified in our previous analysis ¹² were selected for further downstream bioinformatic analyses, including target prediction and pathway enrichment, to elucidate their potential biological functions and mechanistic relevance.

Differentially expressed miRNAs and functional pathway analysis

The confirmed interactions between miRNAs and their target mRNAs were obtained using miRTarBase ¹⁵ , mirTargetlink ¹⁶ , and TargetScan ¹⁷ databases. This tool retrieves data on miRNA–target interactions, combining both computationally predicted and experimentally validated information. We identified the miRNAs from the study in these databases, considering them as candidate miRNAs in conjunction with the mRNAs ¹⁸ . The significantly correlated pairs of these miRNA–mRNA interactions were employed to construct a co‐expression network using Cytoscape 3.6.1. The cytoHubba ¹⁹ v.0.1 plug‐in of Cytoscape ²⁰ was used to select potential hub genes from the identified DE‐miRNAs ²¹ . To identify key regulatory miRNAs, a bottleneck centrality analysis was performed on the constructed miRNA–mRNA interaction network to determine highly influential nodes within the regulatory architecture. miRNAs exhibiting high bottleneck centrality were considered potential driver regulators due to their capacity to control multiple downstream target genes and bridge critical subnetworks. The biological relevance of these candidate miRNAs was further assessed through pathway enrichment analysis using MIENTURNE ²² and DAVID databases ²³ . Gene Ontology (GO) enrichment results were visualized using the ggplot2 R package ²⁴ .

Expression of key miRNAs across tissues in T2D

The five key miRNAs identified through network centrality analysis were evaluated across urinary ECVs and independent transcriptomic datasets from pancreatic islets, blood, liver, and adipose tissue to assess cross‐tissue consistency in T2D. Publicly available datasets were retrieved from the NCBI Gene Expression Omnibus (GEO) database, including pancreas‐derived miRNA dataset (GSE196797) ²⁵ , blood‐derived miRNA data (GSE26168, GPL1032) ²⁶ , liver miRNA data (GSE176025) ²⁷ , and adipose tissue miRNA data (GSE45159) ²⁸ . All datasets included samples from individuals with T2D and corresponding healthy control groups as defined in the original studies.

Cross‐tissue validation of target genes regulated by key miRNAs

Following the identification of key regulatory miRNAs through bottleneck centrality analysis of the miRNA–mRNA interaction network, we examined the expression of their predicted target genes across metabolically relevant tissues. Target genes were derived from the previously established interaction network and restricted to high‐confidence overlapping predictions. To evaluate cross‐tissue expression patterns, publicly available mRNA expression datasets were retrieved from the NCBI Gene Expression Omnibus (GEO), including pancreatic islet data (GSE20966) ²⁸ , blood data (GSE26168, GPL6883) ²⁶ , liver data (GSE130970; including MASH and healthy samples) ²⁹ , and adipose tissue data (GSE78721) ³⁰ . These datasets included samples from individuals with T2D and corresponding healthy controls, as defined in the original studies.

Statistical analysis

All publicly available GEO datasets were downloaded and processed according to platform type. Microarray datasets were background corrected and normalized using the Robust Multi‐array Average (RMA) method, while RNA‐sequencing datasets were normalized using Trimmed Mean of M‐values (TMM) or Transcripts Per Million (TPM), as appropriate. Differential expression analysis between T2D and healthy groups was performed using the limma package ³¹ for microarray data and DESeq2 or edgeR for RNA‐seq data. Statistical significance for transcriptomic analyses was defined as an adjusted false discovery rate (FDR) < 0.05 with |log2 fold change| ≥ 1. Normalized expression values were log2‐transformed for visualization. For urinary ECV miRNA expression analyses, statistical comparisons were performed using PRISM® Version 8.0.2. To account for multiple comparisons, the Bonferroni correction was applied, with significance set at P < 0.01, confirming the relevance of the shortlisted markers. To evaluate the combined diagnostic performance of the miRNA panel, supervised machine learning models including generalized linear model (GLM), support vector machine (SVM), and random forest (RF) were constructed using the caret R package. Models were trained using 10‐fold cross‐validation to reduce overfitting, and classification accuracy was assessed using confusion matrices. Receiver operating characteristic (ROC) curves were generated for each model to compare discriminatory performance. Internal validation of the combined miRNA panel was performed using repeated 10‐fold cross‐validation within the discovery cohort. Predicted probabilities from the optimized model were used to generate ROC curves and calculate the area under the curve (AUC) using the pROC R package ³² . AUC values were used to estimate diagnostic sensitivity and specificity. The Caret R package ³³ was employed to integrate the diagnostic potential of the selected miRNAs. To assess generalizability, the final model trained in the discovery cohort was applied without retraining to an independent external cohort. Predicted probabilities were generated and used to construct ROC curves and calculate AUC values, which were compared with those of individual miRNAs to evaluate robustness and reproducibility. Notably, the external cohort comprised only T2D individuals and lacked an independent control group. Accordingly, ROC analysis in this setting reflects the distribution of model‐derived probabilities rather than a true case–control classification, and should be interpreted with caution. Data visualization was carried out using the ggplot2 R package ²⁴ .

RESULTS

Expression profiles of miRNAs isolated from urinary ECVs of T2D patients

The miRNA expression profiles revealed a clear distinction between the T2D patients and healthy individuals, indicating substantial variability in miRNA expressions associated with T2D status (Figure 1a). The heatmap hierarchical clustering of the top differentially expressed miRNAs further highlighted distinct expression patterns, with several miRNAs showing marked upregulation in T2D patients and others predominantly expressed in healthy individuals (Figure 1b). The volcano plot highlights the significantly dysregulated miRNAs, with several miRNAs (such as hsa‐miR‐671‐5p, hsa‐miR‐185‐5p, hsa‐miR‐302a, hsa‐miR‐320, and hsa‐miR‐204‐3p) significantly upregulated, while others (such as hsa‐miR‐205‐5p, hsa‐miR‐125b‐5p, hsa‐miR‐26a‐5p, hsa‐miR‐203, and hsa‐miR‐107) were significantly downregulated in T2D (Figure 1c).

Figure 1.

Urinary ECVs miRNA profiling in type 2 diabetes (T2D) and healthy individuals. (a) Principal component analysis (PCA) of miRNA expression reveals distinct clustering of T2D (red) and healthy (green) samples. (b) Heatmap of differentially expressed miRNAs showing distinct expression patterns between groups, with red indicating upregulation and blue indicating downregulation. (c) Volcano plot displaying significantly upregulated (red) and downregulated (blue) miRNAs in T2D compared to healthy controls. Selected miRNAs with the most significant changes are labeled.

Dysregulated miRNAs in urinary ECVs reveal disease signatures in type 2 diabetes

Differential expression analysis revealed that 46 miRNAs (19 upregulated and 27 downregulated) were significantly differentially expressed in the urinary ECVs of T2D patients compared to healthy controls (Figure 2a). hsa‐miR‐1246, hsa‐miR‐671‐5p, hsa‐miR‐320c, and hsa‐miR‐155‐5p exhibited the highest upregulation, whereas hsa‐miR‐205‐5p, hsa‐miR‐125b‐5p, and hsa‐miR‐200c‐3p were among the most downregulated miRNAs (Figure 2a). A total of 120 overlapping genes were identified as common targets across all three databases (miRTarBase, TargetScan, and miRTargetLink), ensuring reliability in downstream analysis (Figure 2b). A detailed miRNA‐mRNA regulatory network was constructed to show how the most dysregulated miRNAs are connected to their predicted target genes (Figure 2c). Several hub miRNAs, including hsa‐miR‐1,246, hsa‐miR‐320c, and hsa‐miR‐155‐5p, appeared to regulate multiple target genes, suggesting potential central roles in T2D‐associated pathophysiology. Functional enrichment analysis of the target genes highlighted several significantly enriched signaling pathways (Figure 2d), including MAPK, PI3K‐Akt, AGE‐RAGE, and insulin signaling pathways, all of which are closely linked to diabetes progression and inflammation. Additionally, biological process enrichment (Figure 2e) revealed involvement in cell proliferation, protein phosphorylation, Wnt signaling, cell cycle regulation, and fibroblast migration, indicating that the dysregulated miRNAs may influence key cellular processes implicated in T2D and its complications.

Figure 2.

Functional analysis of dysregulated urinary ECVs miRNAs in T2D: (a) Bar plot showing fold change of significant upregulated and downregulated miRNAs in T2D patients. (b) Venn diagram representing common target genes predicted by miRTarBase, TargetScan, and miRTargetLink databases. (c) miRNA–mRNA interaction network of 46 dysregulated miRNAs (red triangles) and their 120 predicted target genes (green circles), revealing potential regulatory hubs. (d) KEGG pathway enrichment analysis of miRNA target genes. (e) Gene Ontology biological process enrichment. The significance of this association is expressed by the probability (bars) that the association between the targets and the pathway is not due to chance (BH‐adjusted P‐value, Fisher's exact test). (f) mRNA‐miRNA core network module of dysregulated key miRNAs and their experimentally validated or predicted target genes. The key miRNA represents red triangular shape nodes, and the driver gene represents green color circular shape nodes. Black arrow shows the up and downregulated expressed miRNAs. (g) Pathway enrichment dot plot illustrating significantly enriched signaling pathways associated with each of the five miRNAs. Circle size represents the gene count; color gradient indicates false discovery rate (FDR).

Identification of key miRNAs based on network centrality

To prioritize functionally relevant regulators, bottleneck centrality analysis was performed on the miRNA–mRNA interaction network constructed from the 46 significantly dysregulated miRNAs. This network topology approach identifies highly influential nodes that act as critical connectors within regulatory pathways. Among the 46 differentially expressed miRNAs, five (miR‐320a, miR‐16‐5p, miR‐125b‐5p, miR‐26a‐5p, and miR‐30c‐5p) exhibited high bottleneck centrality scores and strong pathway enrichment relevance and were therefore selected as key candidate miRNAs for downstream validation and diagnostic modeling (Figure 2f). A refined interaction network demonstrated that these key miRNAs regulate core metabolic and inflammatory genes, including PTEN, IGF1R, HMGA1, VEGFA, MCL1, CCND2, BTG2, and SMAD4 (Figure 2f). Pathway enrichment analysis confirmed significant involvement of the five miRNAs in VEGF, PI3K‐Akt, MAPK, insulin signaling, AGE‐RAGE, and Wnt signaling pathways (Figure 2g), supporting their central role in T2D‐associated molecular dysregulation.

Diagnostic evaluation and validation of urinary ECV miRNAs in T2D

We next evaluated the diagnostic potential of the five key miRNAs identified from the 46 significantly dysregulated urinary ECV miRNAs. Comparative expression analysis demonstrated clear differences between healthy controls and individuals with T2D (Figure 3a). miR‐320a was significantly increased in T2D, whereas miR‐16‐5p, miR‐26a‐5p, miR‐125b‐5p, and miR‐30c‐5p were significantly reduced. These consistent expression differences suggest their potential as candidate diagnostic markers. To explore clinical relevance, miRNA expression levels were correlated with HbA1c values (Figure 3b). miR‐320a showed a positive correlation with HbA1c, although this association was not statistically significant (Spearman R = 0.13, P = 0.29). Whereas, miR‐16‐5p (R = −0.23, P = 0.059), miR‐26a‐5p (R = −0.39, P = 0.001), miR‐125b‐5p (R = −0.3, P = 0.015), and miR‐30c‐5p (R = −0.27, P = 0.02) demonstrated negative correlations with HbA1c, with miR‐125b‐5p, miR‐26a‐5p, and miR‐30c‐5p reaching statistical significance.

Figure 3.

Diagnostic key miRNA and validation of ECV miRNA in T2D. (a) Box plots comparing the expression levels (log2‐transformed) of selected miRNAs between healthy controls and T2D patients. P values: **P < 0.01, ***P < 0.001. (b) Correlation of key urinary ECV miRNAs expression with HbA1c levels. Spearman correlation coefficients (R) and P‐values are shown in each panel. (d) ROC curves for individual key urinary ECV miRNAs in the discovery cohort. Confusion matrices for GLM, SVM, and Random Forest (RF) classifiers. The X‐axis represents the predicted class, and the Y‐axis represents the actual (reference) class. (e) ROC comparison across machine learning models. (f) Internal cross‐validation ROC comparing individual miRNAs and the combined miRNA panel. (G) Independent external validation ROC evaluating model performance in a separate cohort.

ROC curve analysis revealed moderate discriminatory ability for individual miRNAs. Among them, miR‐30c‐5p showed the strongest performance (AUC = 0.81), followed by miR‐125b‐5p (AUC = 0.79) and miR‐26a‐5p (AUC = 0.76). miR‐320a and miR‐16‐5p also demonstrated reasonable classification performance (Figure 3c). To determine whether combining miRNAs would improve diagnostic accuracy, we constructed GLM, SVM, and Random Forest classifiers. As shown in Figure 3d, GLM and SVM achieved the highest overall accuracy (0.83), while RF showed comparable performance (0.81). ROC comparison across models confirmed strong classification ability, with SVM achieving the highest AUC (Figure 3e). We next evaluated whether integrating the five miRNAs enhanced diagnostic performance. Internal cross‐validation demonstrated that the combined model outperformed individual miRNAs, achieving an AUC of 0.87 (Figure 3f). Importantly, when applied unchanged to an independent external cohort, the combined model maintained strong performance (AUC = 0.86), consistently exceeding that of single miRNAs (Figure 3g).

Validation of urinary ECVs miRNAs dysregulation signature across metabolic tissues in T2D

To assess whether urinary extracellular vesicle (ECV) miRNA dysregulation reflects broader molecular alterations in T2D, we compared the 46 differentially expressed urinary miRNAs with their expression profiles in pancreatic islets, blood, liver, and adipose tissue. Directional comparison across tissues (Figure 4a) revealed partial but not uniform concordance. Several miRNAs demonstrated consistent regulatory trends, with the strongest overlap observed between urinary ECVs and pancreatic islets. Notably, members of the miR‐320 family (miR‐320a, miR‐320b, and miR‐320c) exhibited coordinated upregulation across multiple tissues. Quantitative validation of selected candidates (Figure 4b) confirmed that miR‐320a was significantly increased in urinary ECVs, pancreatic islets, blood, and liver, with no significant change in adipose tissue. In contrast, miR‐16‐5p was significantly reduced in urinary ECVs and pancreatic islets but significantly elevated in blood, indicating tissue‐specific regulation. miR‐26a‐5p showed significant downregulation in urinary ECVs and pancreatic islets, while liver and adipose tissue demonstrated modest increases and blood showed no significant difference. miR‐125b‐5p was significantly decreased in urinary ECVs with largely non‐significant differences across other tissues. miR‐30c‐5p was significantly reduced in urinary ECVs but increased in pancreatic islets, with no detectable differences in blood, liver, or adipose tissue.

Figure 4.

Cross‐tissue validation of urinary ECV‐derived key miRNAs in T2D. (a) Heatmap summarizing the direction of dysregulation of 46 urinary extracellular vesicle (ECV) miRNAs across five biological compartments: urinary ECVs (present study), pancreatic islets (GSE196797), blood (GSE26168), liver (GSE176025), and adipose tissue (GSE45159). Expression status is shown for T2D relative to healthy controls. Purple indicates upregulation in T2D, yellow indicates downregulation in T2D, and black denotes no significant change based on the applied statistical threshold. (H). Purple denotes upregulation in T2D, yellow denotes downregulation in T2D, and black indicates no significant change based on the applied statistical threshold. (b) Box plots displaying log2‐transformed expression levels of selected miRNAs across tissues in healthy controls (H) and individuals with T2D. Boxes represent the interquartile range (IQR) with median; whiskers indicate range. Statistical significance is indicated as *P < 0.05, **P < 0.01, ***P < 0.001; ns, not significant.

Cross‐tissue expression of predicted miRNA target genes in T2D

To determine whether dysregulated urinary ECV miRNAs are associated with coordinated alterations in their predicted target genes, we examined expression of PTEN, IGF1R, MCL1, CCND2, HMGA1, BTG2, VEGFA, and SMAD4 across metabolic tissues (Figure 5). In pancreatic islets, PTEN and BTG2 were significantly upregulated in T2D, while IGF1R, MCL1, CCND2, HMGA1, VEGFA, and SMAD4 showed no significant differences. In blood, MCL1, HMGA1, and VEGFA were significantly increased, whereas other targets remained unchanged. The liver exhibited the most extensive transcriptional alterations, with significant upregulation of PTEN, MCL1, CCND2, BTG2, HMGA1, and VEGFA, alongside significant downregulation of IGF1R. In adipose tissue, PTEN, CCND2, SMAD4, and BTG2 were significantly elevated, whereas IGF1R, HMGA1, and VEGFA were reduced. Overall, target gene expression demonstrated pronounced tissue specificity, with the most robust changes observed in liver and adipose tissue, while pancreatic islets and blood showed more selective alterations, indicating compartment‐dependent regulatory remodeling in T2D.

Figure 5.

Cross‐tissue expression analysis of key miRNA target genes in T2D: Box plots show log2‐transformed mRNA expression levels of validated target genes (PTEN, IGF1R, HMGA1, VEGFA, MCL1, CCND2, BTG2, and SMAD4) in pancreatic islets (GSE20966), blood (GSE26168, GPL6883), liver (GSE130970; MASH and healthy [H]), and adipose tissue (GSE78721). Public datasets were obtained from the NCBI Gene Expression Omnibus (GEO). Statistical significance between healthy and T2D groups is indicated as *P < 0.05, **P < 0.01, ***P < 0.001; ns = not significant.

DISCUSSION

In this study, we showed that urinary ECVs miRNAs are significantly dysregulated in individuals with T2D and, importantly, reflect molecular changes occurring in pancreatic islets while capturing tissue‐specific regulatory heterogeneity in T2D. By integrating small RNA sequencing with cross‐tissue validation and network‐based analyses, we identified key miRNAs that are connected to core pathways underlying T2D, including insulin signaling, PI3K–Akt, MAPK, and AGE‐RAGE signaling. Notably, the combined miRNA signature from urinary ECVs demonstrated reproducible diagnostic performance across both internal and external cohorts. Together, these findings support the concept that urinary ECVs miRNAs can serve as minimally invasive indicators of systemic molecular remodeling in T2D, offering insight into disease biology while also holding promise for clinical application.

Five miRNAs were prioritized from the 46 significantly dysregulated candidates identified between healthy controls and T2D urinary ECV samples using network bottleneck centrality analysis (Figure 2f). Rather than selecting miRNAs solely based on fold change, we applied a systems‐level approach to identify those occupying central and influential positions within key regulatory networks. These networks included PI3K‐Akt, MAPK, insulin signaling, AGE‐RAGE, and Wnt signaling pathways, core metabolic pathways that are well‐established in insulin resistance, inflammation, and β‐cell stress. By using this strategy, the selected miRNAs function not merely as descriptive biomarkers but as biologically meaningful regulators that provide pathophysiological insight into disease mechanisms.

Among the five key miRNAs that showed consistent upregulation across urinary ECVs, pancreatic islets, blood, and liver was miR‐320a, which aligns with its previously described role in pancreatic β cells dysfunction and impaired insulin signaling (Figure 4) ³⁴ . On the other hand, miR‐26a‐5p and miR‐125b‐5p exhibited a distinct pattern, as both were significantly reduced in T2D urinary ECVs and pancreatic islets. This reduction may suggest suppression of regulatory mechanisms that normally protect β‐cell function. This interpretation is supported by previous findings showing that upregulation of miR‐26a‐5p attenuates inflammation in diabetic kidney disease ³⁵ , while miR‐125b‐5p has been demonstrated to improve pancreatic β‐cell function in a T2D mouse model ³⁶ . These directional consistencies add an additional layer of confidence to the biological plausibility of our findings, supporting the notion that urinary ECV miRNAs are not random signals but structured reflections of metabolic disease remodeling.

In order to better understand the biological relevance of our findings, we compared the urinary ECVs miRNA T2D signatures with expression patterns across key T2D metabolic tissues, including pancreatic islets, blood, liver, and adipose tissue (Figure 4). We performed this step to gain insight on whether the captured urinary signals reflect generalized systemic changes or are linked to specific disease‐relevant compartments. Overall, the concordance was strongest between urinary ECVs and pancreatic islets, suggesting that urinary ECVs may capture molecular signals associated with endocrine pancreatic stress. In contrast, other tissues showed more variable patterns, consistent with compartment‐specific adaptation rather than uniform systemic regulation. We think that this observed divergence is biologically informative as it indicates that while pancreatic stress may be reflected at the miRNA level in urinary ECVs, peripheral tissues such as liver and adipose appear to undergo broader transcriptional remodeling, particularly at the mRNA level. Such compartment‐dependent regulation aligns with the distinct yet interconnected roles of these metabolic tissues in insulin resistance and glucose homeostasis.

To uncover key driver genes, we integrated significantly correlated miRNA–mRNA interactions into a regulatory network and used centrality‐based prioritization to identify highly connected hub genes most likely to influence disease‐relevant pathways. Among these genes, the most compelling links to T2D cluster around its two fundamental pathological drivers: impaired insulin signaling (insulin resistance) and progressive β‐cell dysfunction and loss. PTEN is a well‐established negative regulator of PI3K–AKT insulin signaling, placing it directly within the core molecular pathway of insulin resistance ³⁷ . While HMGA1 further connects to this axis through transcriptional regulation of insulin signaling components, with several studies linking its variants or dysregulation to insulin resistance ³⁸ . The second cluster reflects β‐cell adaptation and survival and included CCND2, MCL1, and VEGFA ³⁹ , ⁴⁰ , ⁴¹ . This observed pattern suggests that urinary ECVs cargo could capture a coordinated signature spanning core Diabetes‐related molecular pathways, which support their relevance as systemic, minimally invasive indicators of T2D pathophysiology. Analysis of predicted target genes across metabolic tissues revealed clear compartment‐specific remodeling rather than uniform dysregulation (Figure 5). The pancreatic islets showed selective changes, while liver and adipose tissue exhibited broader alterations in genes central to insulin signaling and metabolic regulation. These findings suggest that dysregulated urinary ECV miRNAs could be linked to coordinated metabolic alterations across organs, reflecting integrated disease remodeling rather than isolated molecular events.

From a diagnostic perspective, the five key dysregulated miRNAs demonstrated robust discriminatory capacity for identifying T2D patients (Figure 3). While individual miRNAs achieved AUC values ranging from 0.73 to 0.81, combining them significantly improved performance (internal AUC 0.87; external AUC 0.86). We are not proposing that this panel replace established diagnostic markers such as HbA1c; however, it may offer complementary value. Its potential role as an early disease biomarker warrants further investigation in larger and longitudinal cohorts. These findings align with emerging evidence positioning ECVs as critical mediators of interorgan metabolic crosstalk and as promising non‐invasive biomarkers in cardiometabolic disease ⁴² .

Despite the promising findings, this study has several limitations. First, the analysis was based on a single population cohort, which may limit generalizability across diverse ethnic groups. Second, the T2D group was significantly older than the healthy controls. Although this reflects real‐world disease demographics and the close biological relationship between age, disease duration, and cumulative metabolic exposure, residual age‐related effects on urinary ECV miRNA expression cannot be completely excluded. Third, detailed medication data were not available for stratified analysis. Therefore, the potential influence of treatment on miRNA profiles cannot be fully disentangled from disease‐related effects. Fourth, although we applied repeated cross‐validation and external validation to reduce overfitting, the relatively small sample size of the discovery cohort, particularly in the context of multiple machine learning models, may still introduce model optimism. Feature selection and model training were performed within the same dataset, which may inflate performance estimates. Therefore, the reported diagnostic accuracy should be interpreted cautiously and requires confirmation in larger, independent cohorts. Finally, the cohort represents clinically established T2D rather than newly diagnosed or treatment‐naïve individuals. Future prospective studies including prediabetic subjects, newly diagnosed patients, and stratification by disease duration will be important to further clarify the intrinsic diagnostic and biological specificity of urinary ECV miRNAs.

In conclusion, our results demonstrate that urinary ECVs miRNAs offer a robust, non‐invasive approach to diagnosing T2D and monitoring its progression. The identified miRNAs, particularly when combined as a panel, exhibit excellent diagnostic accuracy and reflect key pathophysiological processes in T2D. Validation across metabolic tissues underscores their systemic relevance, while correlations with HbA1c highlight their potential as indicators of glycemic control. Future studies should explore longitudinal changes in these miRNAs and their therapeutic implications, potentially paving the way for personalized management of T2D.

DISCLOSURE

The authors declare no conflict of interest.

Approval of the research protocol: The study protocol and procedures were reviewed and approved by the Ethical Review Committee of Dasman Diabetes Institute (Protocol No. RA HM 2019‐008) and conducted in accordance with the principles outlined in the Declaration of Helsinki.

Informed consent: All participants provided written informed consent prior to their enrollment in the study.

Approval date of Registry and the Registration No. of the study/trial: N/A.

Animal Studies: N/A.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.