Circulating Small Noncoding RNA Profiling as a Potential Biomarker of Atherosclerotic Plaque Composition in Type 1 Diabetes

Patients with long-standing type 1 diabetes (T1D) are more likely to progress to macrovascular disease. Yet, many will remain free of atherothrombotic events (1). Although optimal metabolic control remains crucial for preventing cardiovascular disease (CVD), other protective factors might be relevant. Patients with T1D show a higher incidence of plaque ruptures caused by a more aggressive atherogenic process with a worse outcome than subjects without diabetes (2): this condition is determined by a higher lipid and inflammatory content and thickness of the fibrous cap. Calcium depots play a crucial role: while a microcalcification pattern (i.e., spotty calcium depots) is associated with the high vulnerability of atherosclerotic plaques, extensive calcification is associated with more stable plaques (3). Therefore, plaque composition is the most critical determinant of its stability.

We and others have demonstrated that individuals with type 2 diabetes and echogenic plaques are at a higher risk of cardiovascular events than those with echolucent plaques (4,5); however, echolucent plaques seem better predictors of incident cerebrovascular events (6). Few studies have investigated carotid plaque composition in individuals with T1D, who are more prone to developing echogenic and extensively calcified plaques than in subjects without diabetes (7). Although ultrasonography is the easiest approach to define plaque structure, the importance of these observations is threefold: 1) there is a consistent heterogeneity in the pathophysiology of plaque formation, composition, and progression; 2) some yet-unknown factors are operative in the development/prevention of macrovascular disease in patients with T1D (8); and 3) there is the need to recognize markers that can identify patients at risk for developing macrovascular complications and the type of atherothrombotic plaque.

miRNAs, the most abundant class of circulating small noncoding RNA (sncRNAs) regulating gene expression, may have an essential role in CVD development (9,10). We have previously identified the potential value of circulating miRNAs as predictive biomarkers, as circulating miR-30c-5p is responsible for the early proinflammatory events in the arterial wall (11).

Therefore, the aims of the current study were 1) to develop a protocol based on next-generation sequencing (NGS) technology for determining circulating sncRNA in patients with T1D; and 2) to determine the association between ultrasound-determined carotid plaque composition and differentially expressed plasma miRNAs and their variants. Our findings demonstrate that circulating miRNAs associate with different plaque compositions in patients with T1D; this may be relevant to plaque formation and progression.

The present cross-sectional study consists of an exploratory and validation phase. In the discovery phase, the profiling of circulating sncRNAs of 51 patients with T1D (exploratory cohort) was blindly investigated using NGS technologies (Fig. 1). After sequencing and quantifying sncRNAs in the samples, we performed the differential expression (DE) analysis among miRNAs and isomiRs (miRNA variants). We divided the study cohort into two groups according to echogenic carotid plaque composition: carotid fibrous plaque (CFP; n = 25) or carotid calcified plaque (CCP; n = 26). Then, we determined the correlations between clinical variables and the expression levels of the candidate miRNAs.

In the validation phase, candidate miRNAs expression was measured in plasma samples of the validation cohort composed of 104 subjects with T1D (exploratory cohort included) with CFP (n = 40) or CCP (n = 40) and without carotid plaques (control [CTR]; n = 24), using digital droplet PCR (ddPCR). To gain insight into the functional role of the miRNAs in vascular calcification, we analyzed their putative gene targets, enriched for pathways implicated in atherosclerosis, vascular calcification, and diabetes, through the gene ontology functional analysis and the gene ontology biological process to finally identify new biomarkers and mediators for atherosclerotic plaque composition in T1D.

This study was performed with the collaboration of the University of Padova (Padova, Italy), University Hospital de la Santa Creu i Sant Pau (Barcelona, Spain), University Hospital Germans Trias i Pujol (Barcelona, Spain), and University Hospital Arnau de Vilanova of Lleida (Catalonia, Spain). Inclusion criteria were: aged >18 years; T1D for at least 1 year; and estimated glomerular filtration rate >60 mL/min/1.73 m². Exclusion criteria were: previous cardiovascular diseases, defined as clinical coronary heart disease, stroke, or peripheral vascular disease (including any form of diabetic foot disease); and a urine albumin-to-creatinine excretion ratio >300 mg/g. Patients were enrolled in the study if they showed a uniformly or predominantly CFP or CCP; the two groups were matched by sex and age.

This study was carried out following the international ethical guidelines and the principles of the Declaration of Helsinki; the Local Ethics Committees of the participating centers approved the protocol (PI11/11, C0000046, and PI-13–095); and all participants signed informed consent. Patients filled out a lifestyle questionnaire regarding medical history, current therapy, and smoking habits. Plasma glucose, total serum cholesterol, HDL-cholesterol, and triglycerides were measured using routine enzymatic methods. For NGS analysis, fasting blood samples were collected into EDTA-containing tubes, centrifuged at 3,000g for 10 min, and plasma aliquots stored at −80°C until use.

The detailed protocol to evaluate the presence of carotid plaques by ultrasound has been previously described (4,7,12,13). Atherosclerotic plaques were defined according to the Mannheim consensus (12) and classified using the five-type classification system based on visual assessment of echogenicity: uniformly echolucent, predominantly echolucent, uniformly echogenic, predominantly echogenic, and extensively calcified plaques (7). We reclassified plaques into two clinical categories: uniformly/predominantly fibrous (hypoechoic, CFP) and uniformly/predominantly calcified (hyperechoic, CCP) (4,7). The arterial territories explored included the common and internal carotid domains and the bifurcation from both carotid arteries.

All participants underwent the same carotid ultrasound examination, and all measures and ultrasound studies were assessed at each participating hospital by the same researcher. A flow diagram showing the selection of subjects is shown in Supplementary Fig. 1.

sncRNAs from plasma samples were extracted using the miRNeasy Serum/Plasma Kit (Qiagen, Hilden, Germany). sncRNA concentration was determined by Qubit microRNA Assay Kits and measured by Qubit 3.0 Fluorimeter (Thermo Fisher Scientific, Waltham, MA). cDNA synthesis and library preparation of sncRNAs were obtained using the QIAseq miRNA Library Kit (Qiagen). Presequencing and postsequencing quality control analyses were performed. sncRNA libraries were analyzed by LabChip GX Touch Nucleic Acid Analyzer (PerkinElmer, Waltham, MA). We obtained typical electropherograms from sncRNA libraries with a peak between 170 and 180 bp corresponding to an miRNA-sized library and a smaller peak of ∼188 bp corresponding to a piwi-associated RNA (piRNA)–sized library.

sncRNA libraries were sequenced using the Illumina MiSeq Platform. Flow Cell with v3 chemistry and 150 cycles (MiSeq reagent kit v3) was used to perform the sequencing in single reads of 75-bp fragments for the sncRNA library. Replicates were included in the study to verify the result accuracy, reproducibility, and batch effect.

The quality of the NGS runs was evaluated by several parameters as cluster density, passing filter cluster percentage, and Q-score. In our experiments, we obtained an optimal cluster density (1,718.68 ± 117.81 K/mm²), with high passing filter cluster percentage (83.89 ± 2.45) among the sequencing runs and Q30 (Q-score) with an average of 95.35% ± 0.75. Finally, the data were collected as FastQ files.

In this study, two different bioinformatics tools were compared to evaluate their sensitivity and accuracy in detecting and quantifying sncRNAs: Partek Flow software (Partek, Chesterfield, MO) and CLC Genomics Workbench 21.0.3 (Qiagen). These bioinformatics tools provide specific pipelines for sncRNA analysis. An sncRNA quantification pipeline was set up with the same parameters and features for both of the tools. A sequential alignment strategy was used to map sequences on several databases: reference GRCh38 human genome was used, miRBase v.22.1 was considered the first annotation model (14), and sequences were collected as mature miRNAs. isomiRs were identified from the known hairpin sequences in miRBase. Sequences that do not match with miRBase annotated sequences were aligned with piRBase v.1 (15) for piRNAs, with MINTbase v.2 for tRNA-derived small RNA (tsRNAs) (16). Data were filtered before normalization and DE analysis, considering sncRNAs with a minimum number of reads count by setting a cutoff (mean read count >3). All sequences were deposited into the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE207901.

Finally, we performed DE analysis using three different statistical models: 1) the package DESeq2 provided by Partek Flow, which internally corrects for library size; 2) the gene-specific analysis (GSA) task in Partek Flow; and 3) generalized linear model (GLM) by CLC Genomics Workbench. For GSA and GLM, trimmed mean of M-values (TMM) normalized counts were used as input data.

Gene targets for candidate miRNAs were identified and compared using a bioinformatics tool with an online target prediction algorithm, miRWalk 3.0 (http://zmf.umm.uni-heidelberg.de/), a suite of 13 existing miRNA-target prediction programs. We selected common gene targets for miRNAs upregulated in T1D with CCP compared with CFP and gene targets for downregulated miR-10a-5p, significantly enriched in the atherosclerosis network. Target genes were selected using gene set enrichment analysis through gene ontology functional analysis and gene ontology biological process; for the enrichment analysis, we implemented the results using Search Tool for the Retrieval Interacting Genes (STRING v11.5) and clustering the network targets on high-throughput text mining. To create the visualization summary networks, Cytoscape v3.9.0 was used.

cDNA synthesis of sncRNAs and ddPCR was performed with TaqMan probes (Thermo Fisher Scientific). Briefly, reverse transcription was accomplished using the TaqMan Advanced miRNA cDNA Synthesis Kit. Then, cDNA was prepared for ddPCR. For each ddPCR assay, diluted cDNA samples were mixed with 10 μL 2× ddPCR Supermix for Probes (Bio-Rad Laboratories, Hercules, CA) and 1 μL 20× TaqMan Advanced miRNA Probes for miR-93-5p, miR-503-5p, miR-10a-5p, miR-106b-3p, and let7d-5p. Droplets were generated by QX200 AutoDG Droplet Digital PCR system (Bio-Rad Laboratories). The cycling conditions were: 95°C for 10 min, 40 cycles of 95°C for 3 s, 60°C for 30 s, and a final step at 98°C for 10 min. At the end of the PCR reaction, droplets were read in the QX200 droplet reader and analyzed using the Quantasoft version 1.7.4 software (Bio-Rad Laboratories). In addition, a no template control was included in every assay. miRNA quantification with ddPCR was performed with small RNA concentration-normalization to minimize variations in eluted RNA concentrations. Expression values of each miRNA are determined as log copy numbers ratio to input RNA (copies per nanogram of input RNA).

For the sncRNA exploratory study, a sample size calculation was performed to assure a statistically significant difference between sncRNAs from plasma of patients with carotid fibrous or calcified plaques. We performed a power calculation to test whether our sample size could provide enough statistical power to detect a fold change of at least 1.4 for the expression level of sncRNAs between the two groups. For this goal, sample size should be at least 19 subjects/group to reach the desired power of 80%, with an SD estimated at 0.5 and an average power with a false discovery rate of 10%. We studied 25 subjects with CFP and 26 subjects with CCP for the NGS analysis, so our sample size should be large enough to reach the required fold change (17).

Continuous variables are expressed as mean ± SE. Student t test or ANOVA was used as appropriate. For association between studied variables, regression analyses were performed. Statistical significance was accepted at P < 0.05. SPSS version 27 (IBM SPSS Statistics, Segrate, Italy) and GraphPad Prism version 9.0.0 (GraphPad Software, San Diego, CA) were used. Receiver operating characteristic analysis was performed using MedCalc version 19.1.5 (MedCalc Software Ltd, Ostend, Belgium).

The data sets generated and/or analyzed during this study are available from the corresponding authors on reasonable request.

The main clinical parameters of the exploratory study cohort of patients with T1D are reported in Supplementary Table 1. Among all parameters, only the level of LDL-cholesterol was significantly increased in patients with CFP compared with patients with CCP, and the percentage of smokers was significantly higher in CCP than in CFP. In the validation study cohort, we categorized 104 patients with T1D into three groups, according to the presence of CCP (n = 40), the presence of CFP (n = 40), or no carotid plaque determined (CTR; n = 24). LDL-cholesterol concentration was still increased in T1D with CFP compared with patients with CCP, while no other significant differences were observed (Supplementary Table 2).

We performed an sncRNome-wide quantification by NGS of 25 individuals with T1D and CFP and 26 with CCP (exploratory study cohort). In our plasma samples, sequences in the typical range size of 15–55 nucleotides were considered sncRNA transcripts. In Fig. 2A, we report an exploratory analysis of the data showing the percentage distribution of the sequences across the noncoding RNAs (ncRNAs): we obtained a consistent enrichment of sncRNA sequences (68%) in our samples; other transcripts mapped to the human genome represent other ncRNAs (11%) or sequences not yet annotated (20%), while 1% of the reads were unmapped molecules with the same size of sncRNAs. The bioinformatics analysis showed that among the 2,632 miRNAs aligned, 1,362 were expressed in both groups at least in 1 sample. After filtered correction, the bioinformatics analysis covered 465 miRNAs (Fig. 2B). A heat map of the 50 most expressed miRNAs is reported in Supplementary Fig. 2.

Then, in the discovery phase, we analyzed the levels of 465 miRNAs in the plasma of subjects with T1D and CFP and CCP. The volcano plot in Fig. 2C shows 45 significantly and differentially expressed miRNAs in CCP compared with CFP. Since the normalization method and the statistical model affect the final results obtained from RNA-sequencing (RNA-seq) data, we performed DE analysis through two bioinformatic tools and three approaches, and a comparative tightening analysis was performed to obtain unbiased results (Fig. 2D). We selected five miRNAs that were consistently significantly and differentially expressed in accordance with all three approaches to be promoted for further analysis: four miRNAs (miR-503-5p, let-7d-5p, miR-106b-3p, miR-93-5p) were upregulated, and miR-10a-5p was downregulated in plasma of CCP in comparison with CFP (Fig. 3A). However, a functional enrichment analysis showing their interaction with those seven miRNAs identified by only two methods is reported in Supplementary Fig. 3.

We then explored correlations between clinical variables and the expression levels of the five selected miRNAs (miR-503-5p, let-7d-5p, miR-106b-3p, miR-93-5p, and miR-10a-5p) in the exploratory cohort. As reported in Fig. 3B, we observed that all of the selected miRNAs show significant correlations with LDL concentrations; we found a negative correlation with the upregulated miRNAs in the presence of calcified plaque and a positive correlation with the downregulated miRNA (miR-10a-5p). Moreover, the total cholesterol level was inversely correlated to the circulating expression values of the upregulated miRNAs miR-93-p, miR-106b-3p, and miR-503-5p in patients with CCP. However, after adjustment for plaque type (CCP vs. CFP), we found a borderline, but no more significant, correlation between LDL-cholesterol and miR-106b-3p (r = 0.269; P = 0.059) and miR-93–5p (r = 0.266; P = 0.061). These results suggest that LDL-cholesterol could influence, at least partially, the expression of circulating miRNAs, representing one of the most relevant factors affecting the plaque composition. Other significant correlations are summarized in Supplementary Table 3.

We then validated the NGS results obtained from the aforementioned unbiased DE analysis by measuring the expression of the selected miRNAs candidates as potential biomarkers of atherosclerotic plaque composition in T1D of the validation cohort. We determined the expression level of miRNAs using ddPCR, because this technique allows generating absolute expression values and discriminates low-expressed molecules. As shown in Fig. 3C, we confirmed the significant upregulation of miR-93-5p, miR-106b-3p, miR-503-5p, and let-7d-5p in patients with CCP compared with patients with CFP, but we could not confirm the downregulated expression of miR-10a-5p.

Moreover, we found that miR-106b-3p and miR-503-5p were significantly upregulated in patients with CCP and CTR compared with patients with CFP (Supplementary Table 4).

In the validation cohort, the upregulated miRNAs in CCP still showed significant negative correlations with lipid profile: let-7d-5p, miR-106b-3p, miR-503-5p, and miR-93-5p inversely correlated with LDL-cholesterol concentrations, miR-93-5p and miR-503-5p with total cholesterol, and triglyceride with miR-106-3p (Supplementary Table 5). Finally, receiver operating characteristic curve analyses were performed to determine the discrimination capacity of the five miRNAs to distinguish the plaque composition in the validation study cohort: miR-503-5p and miR-106b-3p indicate the best performance to distinguish CFP versus CCP and CTR (Supplementary Table 6).

Next, we analyzed the expression levels of isomiRs, miRNA variants, to substantiate whether specific isoforms could further differentiate the plaque composition in T1D. The bioinformatics analysis considered 562 isomiRs (Supplementary Fig. 4): a heat map of the 50 most expressed isomiRs is shown in Supplementary Fig. 5. We then compared T1D with different plaque compositions: 12 isomiRs were significantly different between the two groups. Among the significant differentially expressed isomiRs, six isoforms were variants of miRNAs already identified as candidate biomarkers for discriminating the plaque composition. Two variants of miR-106b-3p (miR-106b-3p.gc and miR-106-3p.TAs.gc) were consistently upregulated in T1D with CCP compared with CFP, while three variants of canonical miR-10a-5p (miR-10a-5 p.g, miR-10a-5p.tg, and miR-10a-5p.ga) were downregulated in CCP (Fig. 3D and Supplementary Fig. 6). The sequences of significant miRNA variants with their canonical sister and correlations between clinical variables and the expression levels of the isomiRs are reported in Supplementary Tables 7 and 8.

To gain insight into the functional role of selected miRNAs in vascular calcification and atherosclerotic plaque composition, we analyzed their potential gene targets enriched in the atherosclerosis network and clustered the network targets on high-throughput text mining. Supplementary Tables 9 and 10 summarize the significantly enriched pathways in putative gene targets of candidate miRNAs, and, using Cytoscape, we created a biological network showing the match of predictive targets for upregulated miRNAs and the downregulated miR-10a-5p in patients with CCP and their primary biological process (Fig. 4). We found that many target genes for upregulated miRNAs in subjects with CCP are implicated in apoptosis, cell differentiation and activation, and inflammation. Gene targets of the downregulated miR-10a-5p in CCP are mainly involved in osteogenic differentiation, cell-to-cell communication, cholesterol transport and efflux, extracellular matrix organization, and protein degradation.

We performed a comprehensive sncRNA analysis by small RNA-seq in plasma from matched samples of patients with T1D, according to the different composition (CCP vs. CFP) of carotid plaques, determined by ultrasound. Our main findings are the following: 1) four miRNAs (miR-503-5p, let-7d-5p, miR-106b-3p, and miR-93-5p) were significantly upregulated, and one, miR-10a-5p, was downregulated in plasma of T1D with CCP compared with patients with CFP; 2) in a larger validation cohort, also including a control group of patients with T1D without carotid plaque, all four upregulated miRNAs were differentially expressed in the two groups, with the strongest discrimination capacity observed for miR-106b-3p and miR-503-5p; and 3) among the 12 variants of miRNA canonical sequences (isomiRs) significantly differentially expressed between the two groups, 6 were isoforms of miRNAs that we already identified as candidate biomarkers for discriminating plaque composition (miR-106a-3p and miR-10a-5p).

To the best of our knowledge, this is the first study that investigated miRNA profiling concerning different carotid plaque echogenicity that could provide not only further understanding of plaque pathophysiology and progression, but also identify novel biomarkers of plaque composition.

Among the five miRNAs selected in the current study, miR-106b-3p and miR-503-5p represent the most promising potential marker for the discrimination of the plaque composition at the carotid level. Overall, many pieces of evidence support the plausibility of our findings: the overexpression of miR-106b-3p plays an essential role in developing atherosclerotic plaque by promoting foam cell formation and inhibiting macrophage cholesterol efflux via ABCA1 (18). miR-106b cluster is upregulated in patients with coronary heart disease compared with patients with stable angina or noncardiac chest pain through the modulation of the TGF-β signaling pathway (19). The miR-106b cluster participates in the atherosclerotic process and exerts an antiangiogenic effect in endothelial cells by directly targeting STAT3 and stabilizing the plaque (20). In contrast, circulating miR-503 was already found to be dramatically increased in patients with diabetes with critical limb ischemia. Furthermore, the overexpression of miR-503-5p in CCP could be relevant to the calcification processes. Chen et al. (21) demonstrated, in vitro, that miR-503 inhibits osteoclast differentiation and bone loss through translational repression of RANK at the transcription level. These results were also confirmed in vivo through the upregulation of miR-503 induced by agomiR-503 administration (21).

Moreover, miR-503 was upregulated in unstable plaques of a mouse model of plaque instability that mimics atherosclerotic plaque development of humans (22). In humans, miR-503 was significantly overexpressed in patients with diabetes and acute ischemic stroke (23); to support this evidence, the inhibition of miR-503 improved postischemic blood flow and neovascularization in diabetic mice (24). Overall, our findings indicate that these miRNAs may represent potential biomarkers of plaque composition.

miR-10a-5p was the only significantly downregulated miRNA in CCP compared to CFP; this reduction was not significant in the validation cohort analysis assessed by ddPCR. However, its isoforms, detected by NGS technologies, were significantly downregulated in subjects with T1D and CCP; this finding suggests their specific contribution in the modulation of proatherogenic targets. In this context, recent evidence supports that 1) miRNA variants have a pathophysiological role as well as their canonical sequences; and 2) isomiR expression varies among different tissues, showing their peculiarity and potential role as biomarkers (25). Indeed, the functional role of miR-10a-5p isomiRs needs further investigation. Also, our data agree with those of Fang et al. (26), who found that miR-10 expression was significantly reduced in the atherosclerotic plaques from animal models.

Interestingly, miR-10a levels were reduced by TNF-α and IL-1β stimulation, suggesting a transcriptional repression of miR-10a during inflammation (27). Zhang et al. (28) demonstrated that miR-10a-5p is a negative regulatory factor during osteoblast differentiation by inhibiting osteogenic differentiation of bone marrow–derived mesenchymal stem cells. Therefore, the decrease of miR-10a-5p expression and its relative variants indicate that this miRNA may reflect a chronic immune inflammatory process related to the development of atherosclerosis in T1D.

To gain further insight into the functional role of selected miRNAs in regulating plaque composition (29,30), we explored possible correlations with clinical variables. The observed correlations between the selected miRNAs with LDL cholesterol may suggest that plasma lipid profile could modulate circulating miRNA involved in plaque composition in T1D. This hypothesis needs further support.

The bioinformatic analysis of miRNA gene target network showed that the identified candidate miRNAs target the expression of several genes that may impact pathways involved in lipid metabolism, vascular remodeling, and inflammation, with high statistical power. In particular, gene targets of upregulated miRNA are mainly related to the apoptotic process, cell communication, and differentiation, while the gene targets of downregulated miRNA are involved in vascular calcification, cholesterol, and vesicle transport (31,32).

Although we set up a protocol for circulating sncRNA using NGS technology, some methodological considerations should be acknowledged. First, several features of miRNA-seq data may influence the results (i.e., discrepancies in the detection methodology and biological variability, mainly due to the normalization method used for the analysis). Second, choosing an optimal strategy to avoid read count variability is difficult because it may affect sensitivity and specificity. To overcome this issue, in this study, we compared different bioinformatic tools to provide consistent results on the DE analysis of miRNAs. We validated this panel of differentially expressed miRNAs by ddPCR system in a larger cohort, and we confirmed the ability of miR-503-5p and miR-106b-3p to detect the different plaque compositions in T1D. This miRNA panel, along with ultrasonography, may be helpful in better predicting not only plaque composition, but also the cardiovascular risk in patients with T1D. In this study, we excluded patients with renal impairment, since this condition foresees different pathophysiological mechanisms in plaque formation and progression (33).

Furthermore, several studies have long since demonstrated a strong association between carotid plaque and coronary calcium presence and incidence, as well as CVD events (34–37), in the general population, in whom plaque phenotype shows a significant correlation between carotid and coronary arteries (38). In T1D, a higher prevalence of calcified plaques has been demonstrated for both carotid and coronary arteries, compared with control subjects (7,39,40); however, the confirmation of concordance between carotid and coronary artery calcification has not been specifically demonstrated in T1D so far. Future studies are needed to extend these findings in other populations.

Acknowledgments. The authors thank the patients, the IGTP-HUGTP and Institut d’Investigació Biomédica de Lleida (B.0000682) Biobanks integrated in the Spanish National Biobanks Network of Instituto de Salud Carlos III (PT17/0015/0045 and PT17/0015/0027, respectively), and Tumor Bank Network of Catalonia for the collaboration. The authors also thank P. Di Battista, Maternal and Child Health Department, University of Padova, Italy, for bioinformatic assistance.

Funding. Funding was provided by Università degli Studi di Padova to G.C. (BIRD191345/19 and DOR2058483/20).

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.M., A.A., G.C., and S.V.d.K. were responsible for study conception and design. A.G., E.C., M.H., E.O., and N.A. contributed to data collection. A.G., E.C., C.F.Z., D.B., M.H., E.O., and N.A. contributed to analysis and interpretation of results. A.G., D.M., A.A., G.C., and S.V.d.K. drafted the manuscript. All authors reviewed the results and approved the final version of the manuscript. A.G. and E.C. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This study was presented in abstract form at Nature Congress: Chicago Science 2019–Epigenetics and Genome Editing, Chicago, IL, 12–14 June 2019 and at the Annual Conference of the Italian Society for the Study of Diabetes, “Diabetes and Atherosclerosis,” Bologna, Italy, 7–8 February 2020.