The Bruneck project, which began as a prospective population-based survey with the objective of examining the pathophysiology and epidemiology of atherosclerosis, has since expanded to encompass major human illnesses, such as diabetes.^[16-18] At the baseline evaluations in 1990, the study population included a sex- and age-stratified random sample of individuals from Bruneck (Bolzano Province, Italy) aged 40–79 years. The first five-year period, from 1990 to the re-evaluation in 1995, saw the deaths or relocation of 63 members of the subgroup. The results showed that the remaining population follow-up was 96.5% (n=822). RNA extractions were performed on serum samples collected from 822 participants as part of the 1995 follow-up. From 2000 to 2005, the follow-up was 100% complete for clinical goals and 91% complete for repeated laboratory testing. The Online Data Supplement contains a detailed description of how to evaluate the ankle brachial index. The Bruneck population, like that of other Western countries, reflects society as a whole in many ways. The average age of the participants was 62.9 years, with 49.9% being female and 9.7% having diabetes. Before the experiment began, each study participant provided written informed consent.

Diabetes was defined by the World Health Organization as having glucose levels of 7 mmol/L (126 mg/dL) at fasting, 11.1 mmol/L (200 mg/dL) during the 2-hour oral glucose tolerance test, or being clinically diagnosed with the condition. The self-reported status of DM was unquestionably validated by reviewing general practitioners’ medical records and Bruneck Hospital files.

RNA was extracted from serum specimens collected during the 1995 follow-up of 822 participants using the miRNA basic kit (Qiagen). TaqMan miRNA Arrays A and B were used to assess expression levels after miRNAs were reverse-transcribed with Megaplex primer pools (Human Pools A v. 2.1 and B v. 2.0). Individual miRNAs’ expression was determined using TaqMan miRNA assays.

In the initial microarray screening, quantitative (q) PCR tests were performed on thirteen miRNAs with a link to DM. TaqMan assays were performed twice. Samples were collected from private clinics in Basrah, Iraq, between October 15 and November 15, 2024. Group 1 included 50 people with manifest diabetes from the Bruneck cohort, while group 2 included 35 people who developed diabetes between 2014 and 2024 (incident DM). Controls were 47 people of similar age and sex who had no history of diabetes and fasting glucose levels of 6.1 mmol/L (110 mg/dL) and 7.7 mmol/L (140 mg/dL), respectively. Finally, we analyzed the levels of miR-122 in 132 individuals. All qPCR findings were standardized to both miR-454 and RNU6b and analyzed as uncorrected Ct values because there were no widely accepted standards. The RNA of short nuclear samples had to meet the following criteria: first, it had to be detectable in all samples; second, it had to have a moderate range of expression levels; and third, it could not be linked to the presence of diabetes. Furthermore, miR-454’s expression profile was found to be uncorrelated with the rest of the microRNAs; the profile was located outside the coexpression module of the complex networks of microRNAs in serum. (Fig. 1).

Data were analyzed using STATA version 10 and SPSS version 26.0. Continuous variables are displayed as dichotomous data, with the median represented as a percentage and numbers. The median fold change is shown in Fig. 2.

In order to determine precise probability values, this study used the nonparametric Wilcoxon test for related samples to compare the miRNA levels of people with incident or prevalent diabetes to similar groups of matched controls. Logistic regression analyses were also performed for data, including loge-transformed expression levels of miRNAs (1 per model), C-reactive protein, body mass indexes, social status, waist-to-hip ratio, physical activity, smoking status, and high sensitivity to account for the potentially confusing lifestyle effect feature and other parameters associated with DM. Hosmer and Leme show the detailed process of creating models.^[19] The inclusion of suitable interaction terms allowed for the calculation of first-order interaction between miRNAs and the previously listed factors, as well as sex and age. None of these theories were statistically significant. Using a linear model, the variations in miR-122 between glucose tolerance groups were compared. All of the above probability values were two-sided.

We have used network inference techniques to evaluate the general expression characteristics of miRNAs in DM. Either the context likelihood of relatedness or the Pearson correlation coefficient was used to examine the similarity in miRNA expression profiles among all possible miRNA pairs.^[20] Nodes in undirected weighted networks represent miRNAs, whereas connections (edges) show how comparable pairs remaining dependent above a specified threshold are. Since the context likelihood of relatedness depends on a reciprocal information metric and doesn’t assume linearity, it offers some flexibility in identifying biological correlations that could otherwise go unnoticed, whereas Pearson correlation coefficients evaluate linear links between features (miRNAs).^[21,22] Pearson correlation coefficients were used to find clusters of similarly expressed miRNAs, whereas context likelihood of relatedness found all nonrandomly associated qPCR-validated miRNA profiles. Context likelihood of relatedness was selected for mi-RNAs that were validated by qPCR since it is more adaptable to nonlinear dynamics of miRNA expressions than Pearson correlation coefficients and performs noticeably better than other network inference techniques in detecting physiologically significant associations, despite the fact that the two methods can occasionally produce results that are similar.^[23,24] A scale-free design, which is a feature of most real-world networks, including biological ones, was first shown by the miRNA coexpression network after the Pearson correlation coefficient threshold was chosen.^[25] To guarantee repeatability, thirty differentially expressed miRNAs were assessed according to their network characteristics rather than the degree of over- or underexpression.^[26] The context likelihood of relatedness thresholds were set to allow for the representation of all thirteen mi-RNAs in the network while minimizing the number of connections between them.

During prescreening, the architecture of the global miRNA coexpression network was considered, as well as the presence or absence of overexpression or underexpression of each miRNA. Topological parameters like clustering coefficient, node degree, and eigenvector centrality were carefully computed for every miRNA. An individual miRNA’s node degrees are determined by the total number of edges that are related to it. The cluster coefficients show the extent to which miRNAs are likely to form groups. Strong relationships with other miRNAs that are also important in the network increase a miRNA’s eigenvector centrality, which is a measure of miRNA significance.

After being purchased from Cambrex, the human umbilical vein endothelial cell (HUVEC) was cultured on gelatin-coated flasks in M199 media supplemented with 1 ng/mL endothelial cell growth factor (Sigma), 3 g/mL endothelial growth supplement from bovine neural tissue (Sigma), 10 U/mL heparin, 1.25 g/mL thymidine, 5% FBS, and 100 g/mL penicillin and streptomycin, as previously documented.^[27,28] For six days, HUVECs with a high glucose concentration (25 mmol/L) were cultivated in complete medium. To counteract the effects of osmotic stress, mannitol was added to the full medium (5 mmol/L glucose) in which HUVECs were grown. After counting the cells on day five, a matching number were seeded onto T75 flasks, which were subsequently incubated for a further day.

Vesicles were cut apart as before.^[29] Before the conditioned media was gathered, HUVECs were briefly lysed in QIAzol reagent following a 24-hour period during which they were denied serum and growth factors. To examine the expression of miRNA, cellular lysate was stored at −20°C. First, the conditioned media was precleared for 10 minutes at 800 g to get rid of floating cells. After 20 minutes of centrifugation at 10,600 rpm, endothelium particles—also referred to as optative bodies—were able to be separated. Small microparticles (less than 1 µm in size) that were shed from endothelial cells were then isolated using a second centrifugation stage that was conducted for two hours at 20,000 and 500 rpm. The same centrifugation methods were used to create vesicles from serum. After being revived in PBS, the separated vesicle was kept at −80°C. The miRNeasy kit was used to extract total RNA, as previously mentioned. A NanoDrop spectrophotometer was used to measure the amount of RNA.

Correlation value (PCC>0.91) in this level, the networks were dominated by a few hubs connected to a large number of loosely connected nodes, as is typical in biological networks. The miRNA network consisted of 1020 coexpression connections edges and 120 miRNAs nodes.

Marker selection was employed to choose the 13 differently expressed miRNAs since it is more repeatable to identify their location in the miRNA coexpression network than to identify individual over or under expressions. ^[26] Thirteen networks of differential miRNAs were topologically central. Out of the many miRNAs with different expression levels, 13 were chosen because they showed a broad range of node degrees, clustering coefficients and eigenvector centrality values. miR-454, which was located outside of each network module, was the only miRNA that was demonstrated to be unconnected to the expression of other miRNAs. This is how it was selected as an extra normalizing control.

qPCR helped to improve the quantification of the 13 topographically distinct miRNAs. Every patient with evident diabetes had their age and sex matched. In diabetics, plasma level of miR-21, miR-24, miR-20b, miR-122, miR-191, miR-15a, miR-197, miR-320, miR-223, miR-486, miR-29b, and miR-150 were all lower; however, miR-28-3p was frequently greater (Fig. 2). Results for expressions level standardized to either RNU6b or miR-454 and non-standardized miRNA levels were in agreement (Fig. 2).

Four miRNAs, including endothelium miR-122, remained significant after controlling for the multiple comparison that were performed (probability value 0.000140). There were significant differences in nine miRNAs standardized to RNU6b between DM patients and controls. However, because individual miRNAs in this situation are highly related rather than independent of one another, the Bonferroni correction is overly conservative. Multivariate analysis revealed that all miRNAs, with the exception of miR-29b, were significantly associated with manifest DM. There was a positive inverse relationship between eleven and miR-28-3p. In both diabetes patients with and without treatment, the results for miRNAs standardized to miR-454 are shown in Fig. 3. In a subsequent run, miR-122 was quantified and compared to miR-454. Once again, miR-122 was a strong predictor of manifest DM, according to logistic regression studies (odd ratios [93% CI] for a 1 SD unit reduction of log-transformed expressions levels of miR-122: 1.98 [1.41–2.78]. Furthermore, plasma levels of miR-122 showed a progressive decline in impaired fasting glucose/impaired glucose tolerance (n=35), normal glucose tolerance (n=47) and manifest diabetes (n=50) across categories. The levels of miRNAs and fasting glucose levels were inversely correlated in both diabetic patients and control subjects (r=−0.175 to −0.369) Plasma samples from hyperglycemic patients aged 8–12 weeks (Fig. 3) were able to independently reproduce the majority of the miRNA changes seen in DM.

Importantly, prior to the onset of DM, certain miRNAs were already altered. Over the course of the ten year follow-up period, 50 people in total developed diabetes (mean interval to diagnosis of DM from 2014 to 2024). Values of miR-29b, miR-15a, miR-223 and miR-122 were significantly lower in patients group individuals, but miR-28-3p was higher in matched control group (Fig. 4).

To find out if miRNAs can accurately differentiate between people with incident or prevalent DM and healthy controls, we broke down miRNAs into principle components (PC). The expression patterns of the five most significant miRNAs (miR-122, miR-15a, miR-28-3p, miR-223, and miR-320) allowed for the accurate diagnosis of 60/85 (73%) DM patients and 40/47 (94%) controls (Fig. 5A,B,C). In comparison to a sample of patients with well-controlled diabetes, the 35 DM cases classified as normal subjects had significantly lower fasting glucose (mean ± SD, 130.0±25.9 mg/dL against 150.7± 50.0 mg/dL, p=0.0047) and HbA1c (mean ± SD, 5.94±0.81% versus 6.54±1.86%, p=0.014) values. The performance of the classifier was not enhanced by the inclusion of more miRNAs(Fig. 5B). Therefore, it may be concluded that there is minimal need for miRNA signature-based classification for these five miRNAs. The miRNA relevance network’s inference further supported the potential use of miRNAs as diabetes diagnostic tools.

The miRNA most commonly linked to diabetes mellitus is miR-122. Angiogenesis, wound healing, and the maintenance of vascular integrity are all regulated by this miRNA. It has previously been demonstrated that endothelial cells and endothelial apoptotic bodies contain a significant concentration of miR-122.^[30] In order to ascertain if hyperglycemia influences this process, the miRNA level of microparticle produced under standard and shed endothelium particle (5 mmol/L) along with elevated (25 mmol/L) glucose concentration was examined using miR-122 release from endothelial cells. Severe hyperglycemia had no effect on cellular miRNA concentrations, but it dramatically decreased the level of miR-122 in endothelial dead cells (Fig. 6A). Other than miR-24, the shedding of other miRNAs remained unchanged. The lower levels of miR-122 in diabetics limited to the plasma particulate fraction (Fig. 6B) are in line with these in vitro studies. Lastly, data from our study sample indicates that a decrease in miR-122 in plasma is linked to both preclinical and overt DM illness.