MicroRNA 146a is associated with diabetic complications in type 1 diabetic patients from the EURODIAB PCS

MicroRNA-146a-5p (miR-146a-5p) is a key regulator of inflammatory processes. Expression of miR-146a-5p is altered in target organs of diabetic complications and deficiency of miR-146a-5p has been implicated in their pathogenesis. We investigated if serum miR-146a-5p levels were independently associated with micro/macrovascular complications of type 1 diabetes (DM1). A nested case–control study from the EURODIAB PCS of 447 DM1 patients was performed. Cases (n = 294) had one or more complications of diabetes, whereas controls (n = 153) did not have any complication. Total RNA was isolated from all subjects and miR-146a-5p levels measured by qPCR. Both the endogenous controls U6 snRNA and the spike (Cel-miR-39) were used to normalize the results. Logistic regression analysis was carried out to investigate the association of miR-146a-5p with diabetes complications. MiR-146a-5p levels were significantly lower in cases [1.15 (0.32–3.34)] compared to controls [1.74 (0.44–6.74) P = 0.039]. Logistic regression analysis showed that levels of miR-146a-5p in the upper quartile were inversely associated with reduced odds ratio (OR) of all complications (OR 0.34 [95% CI 0.14–0.76]) and particularly with cardiovascular diseases (CVD) (OR 0.31 [95% CI 0.11–0.84]) and diabetic retinopathy (OR 0.40 [95% CI 0.16–0.99]), independently of age, sex, diabetes duration, A1c, hypertension, AER, eGFR, NT-proBNP, and TNF-α. In this large cohort of DM1 patients, we reported an inverse and independent association of miR-146a-5p with diabetes chronic complications and in particular with CVD and retinopathy, suggesting that miR-146a-5p may be a novel candidate biomarker of DM1 complications.

Whether serum miR-146a-5p levels are associated with vascular complications of DM1 diabetes is unknown; however, miR-146a-5p was one of the 25 differentially expressed miRNAs in a profiling analysis performed in pooled serum samples from DM1 patients with and without complications [20].
The aim of the present study was to confirm these profiling results and to assess if elevated miR-146a-5p levels affected odds ratios (ORs) of micro/macrovascular complications in DM1 patients from the EURODIAB PCS nested case-control study.

Patient sample
The EURODIAB Type 1 Diabetes Complications Study (1989)(1990)(1991) was designed to discover risk factors of diabetes complications in 3,250 subjects with DM1 [21]. Seven years later study participants were invited for re-examination in the EURODIAB Prospective Complication Study (PCS 1997(PCS -1999 [22] and a nested casecontrol study was designed at follow-up. Cases had at least one complication among cardiovascular diseases (CVD), retinopathy, and albuminuria; whereas controls were free of complications [23][24][25][26]. The present study included 460 subjects (300 cases and 160 controls, ratio 1.9) with available serum samples and data on complications. Cases and controls were unmatched and adjustments were made at the analysis stage. Risk factors [cholesterol, hypertension, body mass index (BMI), triglycerides, A1c] were assessed as previously described [26]. Retinopathy was evaluated and graded from retinal photographs using the EURODIAB protocol [22]. Albumin excretion rate (AER) was measured on two 24-h urine collections by immunoturbidimetry and classified as normo-(< 20 μg/min), micro-(20-200 μg/min), or macro-albuminuria (≥ 200 μg/min). Glomerular filtration rate (eGFR) was estimated using the four-component equation from the Modification of Diet in Renal Disease study [27]. CVD was defined as a physician-diagnosed positive history of myocardial infarction, coronary artery bypass graft surgery, angina, stroke and/or ischemic changes on a centrally Minnesota-coded EKG.
The study was approved by the Ethical Committee of Turin University. The procedures were in accordance with the Helsinki Declaration. Written consent has been obtained from each patient or subject after full explanation of the purpose and nature of all procedures used.

RNA isolation
Total RNA was prepared using the Trizol ® LS reagent (Thermo Fisher, Milan, Italy) according to the manufacturer's recommendations. In brief, 200 µl of serum was mixed thoroughly with 750 µl of TRIZOL ® LS reagent. Mixtures were then gently inverted 5-8 times, and incubated at room temperature (RT) for 15 min. After which, both the spike-in Cel-miR-39 (3 µl) and 200 µl of chloroform were added and the solution was mixed vigorously. The samples were then centrifuged at 12,000xg for 15 min at 4 °C, the upper aqueous phase was carefully transferred to a new tube, upon which 500 µl isopropanol was added, and then incubated for 10 min at RT, before then being centrifuged a 4 °C, 12,000xg for 10 min. Pellets were washed with 75% ethanol, air-dried at RT for 10 min, and re-suspended in 25 μl of nuclease free H 2 O. RNA quality was assessed by capillary electrophoresis on an Agilent-2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA).

Statistical analyses
Normally distributed variables are reported as means (standard deviation, SD), while non-normally distributed variables were log-transformed (miR-146a-5p, triglycerides, NT-proBNP, TNF-α, IL6, AER) and reported as geometric means (25°-75° percentiles). Student's t-test and ANOVA were used for comparisons. Categorical variables were compared using the Chi-Squared test. Pearson's correlation coefficient analysis was used to assess the relationship between miR-146a-5p and clinical variables. Logistic regression analysis was employed to estimate the odds ratios (ORs) of miR-146a-5p for all chronic complications (micro-macroalbuminuria, CVD, retinopathy), independently of established risk factors and confounders. We used the likelihood ratio test to compare nested models, exploring the relevance of age, gender, diabetes duration, BMI, A1c, blood pressure, total cholesterol, AER, eGFR, NT-proBNP, IL-6, TNF-α, and smoking. Analyses were hypothesis-oriented and variables were retained in the final model if they added significantly to the likelihood of models or to the estimated coefficients of predictors. As miR-146a-5p may play a different role in various complications, models were also fitted for each complication separately. To assess ORs for increasing miR-146a-5p levels, miR-146a-5p levels were categorized by quartile distribution in controls. Because ORs in the three lower quartiles were similar, they were combined as the reference category in the final analysis and compared to the upper quartile. A p value < 0.05 was considered statistically significant. Analyses were performed using the SPSS (Version 27) software.

Patients' clinical characteristics
The studied population (n = 447) ( Fig. 1 Table 1 shows the characteristics of both cases and controls. Overall cases had a worse risk factor profile than controls, as shown by more unfavorable levels of conventional risk factors and inflammatory cytokines.

miR-146a-5p expression
MiR-146a-5p was measurable in all the 447 samples and value distribution was right-skewed. Individual Ct values are reported in the Additional file 1: Table S1. There were no differences in mir146a-5p levels between men and women

Correlation of miR-146a-5p with clinical parameters and serum biomarkers
We then explored the association between miR-146a-5p and a range of relevant biochemical and clinical variables. Correlations of all variables in a matrix are reported in Additional file 2: Table S2. There was no correlation between miR-146a-5p and age even after adjustment for sex (p = 0.823). Similarly, no correlation was found between miR-146a-5p levels and diabetes duration neither in the whole population nor in cases (cases: r = 0.038, p = 0.521). Notably, serum miR-146a-5p levels were inversely correlated with diastolic blood pressure (r = -0.11, p = 0.019), triglycerides

Logistic regression
Logistic regression analyses were carried out to establish if miR-146a-5p changed the odds ratios (OR) of having either all or individual vascular complications, independently of risk factors and potential confounders. Levels of miR-146a-5p in the upper quartile (≥ 6.73) were associated with a 59% lower risk of having all complications (OR = 0.41, 95% CI 0.21-0.79) and this was independent of age, gender, A1c, and duration of diabetes (Model 1- Table 2). After adjustment for AER, hypertension, eGFR, smoking, the strength of this inverse association was even greater (OR = 0.32, 95% CI 0.14-0.73) (Model 2- Table 2). Following further adjustment for TNF-α and NT-proBNP, miR-146a-5p values in the upper quartile are still reduced by 66% the OR for all complications (OR = 0.34, 95% CI 0.14-0.76) (Model 3- Table 2). Models were not affected by the inclusion of total cholesterol, BMI and IL-6. Analysis performed separately in men and women showed that miR-146a-5p was inversely and independently associated with all complications in men. An inverse association was also observed in women, but did not reach statistical significance (Table 3). We also examined the associations between miR-146a-5p levels and individual diabetes complications. As shown in Table 2, results obtained in all cases were predominantly driven by cases with CVD and retinopathy. In Model 1, levels of miR-146a-5p in the upper quartile were associated with 62% and 59% reduced OR for CVD and retinopathy, respectively. The trend of OR changes across models was similar to that observed for all complications and after adjustment for age, sex, A1c, diabetes duration, AER, hypertension, eGFR, and smoking  the risk reduction for CVD and retinopathy was 68% and 61%, respectively. After further adjustment for both TNF-α and NT-proBNP the strength of these associations remained statistically significant.

Discussion
In this case-control study on DM1 patients from the EURODIAB PCS, we showed an independent and inverse association between miR-146a-5p and chronic complications of diabetes.
Values of miR-146a-5p were lower in cases compared to controls. Moreover, in logistic regression analysis, miR-146a-5p levels in the highest quartile distribution of control subjects were associated with a significant risk reduction of diabetic complications, independently of confounders and established risk factors. This suggests that miR-146a-5p may be a candidate marker of vascular protection in DM1. miRNA-based biomarker efficacy may differ between sexes and previous studies reported lower circulating miR-146a-5p levels in women than in men [28,29] and an influence of sex adjustment on the relationship between circulating miR-146a-5p levels and clinical outcomes [30,31]. In our study, miR-146a-5p levels did not differ in men and women and adjustment for sex did not modify the results. However, when analyses were performed separately in men and women, the inverse relation between miR-146a-5p levels and diabetes complications was confirmed in men, while it did not reach statistical significance in women. The reason for this gender-specific difference is unknown; however, it may be related to the confounding effect of miR-146a-5p negative modulation by estrogens [32,33].
Previous studies have reported an age-related decline in circulating miR-146a-5p levels [28,34]. However, in our study, miR-146a-5p levels did not correlate with age and did not differ across age categories. Moreover, adjustment for age did not modify the results. The reason for this discrepancy is unknown; however, the effect of age on miR-146a-5p levels was observed in very old subjects (> 75 years) [34] and our DM1 patients were relatively young. Moreover, the age-related decline in miR-146 was weaker [34] or even absent [35] in studies performed on patients with type 2 diabetes.
The source of serum miR-146a-5p was not established in our study; however, endothelial cells, platelets, and immune/inflammatory cells are likely candidates [36,37]. In these cell types, miR-146a-5p expression can be modulated by diabetes. Indeed, exposure of endothelial cells to high glucose induces a persistent miR-146a-5p downregulation [38][39][40]. Moreover, diabetes induces an inflammatory response that can trigger a compensatory expression of miR-146a-5p. Therefore, the relationship between miR-146a-5p and diabetes is complex and is likely affected by many other variables besides hyperglycemia. This may explain why in our study there was no correlation between miR-146a-5p and A1c, despite the inverse association between miR-146a-5p and diabetes complications.
The biological mechanism underlying the independent and inverse association between serum miR-146a-5p levels and DM1 complications is unknown. However, low serum miR-146a-5p level may mirror a miR-146a-5p downregulation in endothelial cells, monocytes, and other cell types relevant to DM1 complications, where miR-146a-5p deficiency impairs the feedback restraints on diabetes-induced inflammation [40], NOX-4-mediated oxidative stress [41], and apoptosis [42]. Therefore, serum miR-146a-5p level may be a circulating marker of the efficacy of the anti-inflammatory response and thus of the susceptibility to develop complications. Consistent with this, miR-146a-5p is a well-established marker of inflammation and is associated with a variety of immune and non-immune inflammatory conditions [14,[43][44][45], including CVD, Alzheimer's disease, and type 2 diabetes [34,[46][47][48][49].
On the other hand, low serum miR-146a-5p levels may also be due to reduced miR-146a-5p content in circulating EVs. As uptake of miRNA containing EVs can induce phenotypic changes in the recipient cells, this may hamper the systemic control of miR-146a-5p on inflammation and thus favor the development of DM1 complications [50,51]. Consistent with this, therapeutic strategies delivering nanoparticles loaded with miR-146a-5p have been successfully tested in various inflammatory conditions [14,[52][53][54].
The magnitude of the association between miR-146a-5p and all complications was greater after adjustment for A1C, diabetes duration, hypertension, AER, eGFR, TNF-α, NT-pro-BNP, suggesting that miR-146a-5p measurement may have an added value over traditional risk factors in identifying patients at enhanced risk of complications. Diabetes-related risk factors induce a low-grade inflammation and they are thus expected to enhance miR-146a-5p expression (13). This may explain their negative confounding effect in the inverse relationship between miR-146a-5p and all complications. Once induced, miR-146a-5p potently inhibits inflammatory processes that contribute to diabetic complication onset [13]. Therefore, higher circulating miR-146a-5p levels may identify the subgroup of patients able to build a more effective miR-146a-5p-mediated anti-inflammatory response and thus less prone to develop complications. Both miR-146 polymorphisms and epigenetic modifications may affect the efficacy of miR-146 negative feedback loop [55]. At variance with our results, a recent study in patients with type 2 diabetes showed that miR-146a-5p levels in circulating CD31 + EV were greater in individuals with vascular complications compared with subjects without complications [37]. However, miR-146a-5p CD31 + EV levels were not adjusted for clinical variables that correlated with miR-146a-5p and differed between the two groups. Moreover, miR-146a-5p was measured on circulating CD31 + EV rather than on serum, making direct comparison [37] between the studies difficult.
Logistic regression analysis carried out for individual complications showed that higher miR-146a-5p levels were associated with a 60% lower OR of diabetic retinopathy. In line with this finding, miR-146a-5p is downregulated in the diabetic retina [38] and intervention strategies that enhance retinal miR-146 levels either by intravitreal injection or overexpression, ameliorate experimental diabetic retinopathy [56,57]. Besides lowering inflammation, high miR-146a-5p expression may also affect retinopathy by downregulating the proangiogenic HIF-1α-ROBO4 pathway [58].
We also found an independent and inverse association between miR-146a-5p and CVD. Mechanistically, miR-146a-5p acts as a brake on proinflammatory NF-kB signaling in both endothelial cells and monocytes and this together with miR-146a-5p anti-oxidative properties [41] may explain the link between miR-146a-5p and atheroprotection. Consistent with this, treatment with miR-146a-5p is beneficial in animal models of atherosclerosis [59,60]. Moreover, a miR-146a-5p polymorphism, resulting in lower mature miR-146a-5p production, confers enhanced risk of coronary artery disease in humans [61].
Besides the potential relevance of miR-146a-5p as a biomarker of CVD in diabetes, serum miR-146a-5p may also have direct vascular protective effects. Indeed, serum EV carrying miR-146a-5p can enter into recipient cells and alter the cell phenotype by releasing miR-146a-5p. Consistent with this, EV-mediated transfer of miR-146a-5p has been demonstrated to modulate inflammation and to improve cardiac function after myocardial infarction [62]. Moreover, in mouse models of atherosclerosis, delivery of miR-146a-5p to the endothelium by injection of either exogenous EV or a free miR-146a-5p mimetic not only ameliorated endothelial inflammation, but also reduced atherosclerotic plaque size [59,63].
MiR-146a-5p has also been involved in the pathogenesis of diabetic nephropathy [15,18,64]. However, we did not observe any significant associations between serum miR-146a-5p levels and this microvascular complication of diabetes. The reason is unknown; however, endothelial cells that are a major source of circulating miR-146a-5p may play a less relevant role in this complication than in CVD and retinopathy.
Our study has several limitations. Although EURO-DIAB is a prospective study, baseline serum samples were not available; therefore, miR-146a-5p could only be measured in follow-up samples. Given the cross-sectional design of the study, our results only demonstrate the presence of an independent association between miR-146a-5p and chronic diabetes, but they do not prove a clinical relevance of miR-146a-5p as a biomarker of DM1 complications. Moreover, the cross-sectional design of the study restricts our ability to investigate causal relationships and to elucidate mechanisms. However, pilot cross-sectional studies on large and well-characterized existing cohorts are an effective manner to identify miRNAs of potential interest that can be then tested in longitudinal studies. The smaller number of controls compared to cases has reduced the power of the analyses; however, the number of controls and cases with individual complications was similar, though multiple comparisons might have caused significant results due to chance. Lack of matching between cases and controls for age represents a limitation of the study as age can affect circulating miR-146a-5p levels [28,34]; however, adjustments for age were made at the analysis stage. The study did not include a control group of non-diabetic patients as the EURODIAB study exclusively recruited DM1 patients; however, the major purpose of the study was to assess if miR-146a-5p was associated with diabetes complications rather than compare circulating miR-146a-5p levels in patients with and without diabetes. Results were not adjusted for residual β-cell function; however, the role of C-peptide as marker/therapy for chronic DM1 complications is still controversial. Moreover, C-peptide is usually detectable in the first decade of DM1 [65] and in our study only 10% of patients had a diabetes duration below 10 years. The possibility of sample degradation over time cannot be ruled out. On the other hand, samples were properly stored and miRNAs are stable in biofluids. Although the TRIZOL method for miRNA extraction may be less sensitive compared to specific miRNA