Diagnostic potential of circular RNAs in carotid atherosclerotic plaque vulnerability

Abstract

Background

Vulnerable carotid atherosclerotic plaques are prone to cap rupture and thrombosis, yet definitive assessment still relies on postoperative histology. We investigated whether circulating circular RNAs (circRNAs) in peripheral blood mononuclear cells (PBMCs) can identify plaque vulnerability preoperatively.

Method

In patients undergoing carotid endarterectomy, plaques were classified as vulnerable or stable, and PBMC expression of candidate circRNAs (circVIRMA, circGRN, circANRIL and CDR1as) and paired linear transcripts was quantified by RT-qPCR. A composite in-plaque_SCORE (IL-1β, IL-6, IL-10, PPARγ mRNAs) characterized intraplaque inflammation. We derived two PBMC metrics: circRNA_SCORE (mean expression of deregulated circRNAs) and circ/linear mRNA_SCORE (mean of circ/linear mRNA expression ratios) and evaluated diagnostic performance by Receiver Operating Characteristic curve (ROC) analysis and logistic regression.

Results

circVIRMA and circGRN were modestly increased in PBMCs from patients with vulnerable plaques, whereas linear host genes were unchanged. CDR1as was markedly upregulated and showed the strongest discrimination among the other circRNAs, similarly to the circRNA_and circRNA/linear mRNA ratio_SCOREs. Indeed, circRNA_SCORE correlated positively with the intraplaque inflammatory in-plaque_SCORE, linking systemic signatures to local plaque biology. In plaque tissue, CDR1 mRNA was reduced while CDR1as was relatively preserved, yielding a higher CDR1as/CDR1 mRNA ratio in vulnerable lesions.

Conclusions

PBMC circRNAs—particularly CDR1as—show promise as noninvasive biomarkers of carotid plaque vulnerability and reflect plaque inflammatory status. Validation in independent cohorts and integration with imaging could enable improved preoperative risk stratification.

Highlights

• •circVIRMA, circGRN and CDR1as increased in PBMCs of vulnerable plaque patients; linear transcripts unchanged.
• •CDR1as(1) was strongly upregulated, and its level and ratio to linear CDR1 best predicted disease among circRNAs tested.
• •CDR1as(1) showed discrimination comparable to circRNA and circRNA/mRNA ratio scores, supporting its biomarker potential.
• •circRNA_SCORE positively correlated with in-plaque inflammatory score, linking systemic and local plaque biology.

1. Introduction

Atherosclerotic plaque formation is described by the “response-to-injury” theory, originally proposed by Virchow and later refined by Ross [1]. Endothelial dysfunction and oxidative stress promote low-density lipoprotein (LDL) oxidation and endothelial activation, leading to increased expression of chemokines such as monocyte chemoattractant protein-1 (MCP-1), and adhesion molecules including vascular cell adhesion molecule-1 (VCAM-1), E-selectin, and P-selectin that drive leukocyte recruitment. Recruited monocytes differentiate into macrophages, ingest oxidized LDL, and become foam cells that release pro-inflammatory cytokines and mediators such as interleukin-6 (IL-6), tumor necrosis factor-alpha (TNF-α), and interleukin-1β (IL-1β), sustaining local inflammation. Smooth muscle cells contribute through phenotypic switching and extracellular matrix deposition, generating a fibrous cap over a lipid-rich necrotic core. Plaque vulnerability is influenced by inflammatory activity at the ‘shoulder region’ and matrix metalloproteinase-mediated cap degradation, which weaken the fibrous cap and increase rupture risk, with overall stability determined by the balance between fibrous cap integrity and lipid core burden [2,3]. Despite medical advances, atherosclerosis remains a major cause of illness and death. Plaque formation and arterial stenosis underlie severe clinical outcomes, including myocardial infarction when affecting the coronary arteries and ischemic stroke when occurring in the carotid arteries. Stroke is particularly significant from an epidemiological point of view, as it is ranking as the second leading cause of death worldwide and third for combined death and disability from non-communicable diseases [4].

The concept of the “vulnerable plaque” was first introduced by Müller et al. [5] in the context of coronary artery disease and was later extended to the carotid arteries in subsequent studies [6]. At present, definitive identification of plaque vulnerability features is only possible following surgical removal of the lesion, through macroscopic evaluation [7] and histological analysis [8]. Vulnerability is most often associated with the rupture of the fibrous cap and subsequent thrombus formation [9]. However, arterial thrombi may also arise in the absence of rupture, as in cases of superficial endothelial denudation, also termed plaque erosion [10,11]. These events highlight the importance of carotid plaque vulnerability in terms of cerebrovascular disease progression. There are several evidence indicating that plaque disruption is strongly associated with both incident and recurrent ischemic stroke, frequently independent of stenosis severity. Specifically, intraplaque hemorrhage (IPH) was highly prevalent (51.6% in symptomatic patients and 29.4% in asymptomatic patients) and was a robust predictor of ipsilateral stroke, conferring increased risk in both symptomatic (Hazard Risk (HR) = 10.2) and asymptomatic individuals (HR = 7.9). Across all stenosis categories, stroke rates were consistently higher in the presence of IPH compared with its absence. Furthermore, both IPH (HR = 11.0) and severe stenosis (HR = 3.3) remained independent predictors of ipsilateral stroke [12]. In an additional large-cohort study, plaque surface irregularity was independently associated with ipsilateral ischemic stroke across all degrees of stenosis [13].

Homburg et al. [14] studied 346 ischemic stroke patients with ≥50% and <50% carotid stenosis and found that plaque ulceration was common in both groups. A higher proportion of lipid-rich necrotic core was strongly associated with ulceration (OR 2.21; 95% CI 1.49–3.27), while greater calcification was linked to lower odds (OR 0.60; 95% CI 0.40–0.89), independent of age, sex, and stenosis severity. These associations remained significant in low-grade stenosis, supporting the value of assessing plaque burden and composition to improve identification of rupture-prone plaques and stroke risk stratification.

At present, the identification of vulnerable carotid plaques is largely based on macroscopic and histopathological examination of the excised plaque. Although these modalities have improved the identification of high-risk morphological features, they remain indirect surrogates of the underlying biological and molecular activities driving plaque instability. Computed tomographic (CT) or Magnetic resonance imaging (MRI) imaging has been proposed as a non-invasive tool for plaque evaluation; however, its clinical use remains limited due to cost, radiation exposure, infrastructure demands, and the requirement for specialized expertise, resulting in restricted accessibility and suboptimal reliability in assessing plaque vulnerability [15,16]. This limitation underscores a major unmet clinical need: the availability of reliable circulating biomarkers capable of identifying vulnerable plaques non-invasively, prior to carotid endarterectomy. Such biomarkers could significantly enhance early risk stratification, guide therapeutic decision-making, and optimize patient selection for surgery. The development of blood-based indicators reflecting intraplaque inflammatory activity would represent a crucial step toward more personalized and timely management of carotid artery disease.

Numerous studies have investigated potential preoperative predictors—clinical, biochemical, and radiological—that might correlate with carotid plaque vulnerability parameters [[17], [18], [19]]. Circulating biomarkers, in particular, have been examined for their association with increased atherosclerotic burden, accelerated disease progression, and features of carotid plaque instability [20,21]. Nevertheless, none of the proposed markers has demonstrated sufficient sensitivity, specificity, and reliability to allow routine use in clinical practice [22,23]. These limitations underscore the importance of composite diagnostic approaches that integrate circulating biomarkers with intravascular imaging for improved risk stratification and personalized therapeutic strategies.

Early research in atherosclerosis mainly examined protein-coding genes, but the development of RNA sequencing has revealed numerous noncoding RNAs (ncRNAs) [24], expanding insights into plaque biology [25,26]. NcRNAs, which are not translated into proteins, are divided by length into small ncRNAs (e.g., microRNAs, miRNA, <200 nucleotides) and long ncRNAs (lncRNA, >200 nucleotides). Circular RNAs (circRNAs), a type of long noncoding RNA, are increasingly recognized for their roles in disease regulation and as potential diagnostic biomarkers. [25,[27], [28], [29], [30]]. Similar to other ncRNAs, circRNAs have properties that make them promising biomarkers [31]. They can be detected in blood samples [27], including plasma, serum, and peripheral blood mononuclear cells (PBMCs)—which consist of T cells, B cells, natural killer cells, and monocytes—which often mirror pathological changes in tissues [32].

circRNAs, miRNAs and mRNAs can interact forming complex regulatory networks. miRNAs suppress gene expression by binding to specific microRNA response elements (MREs) within target transcripts [33]. When circRNAs and mRNAs harbor shared MREs for the same miRNA, they may reciprocally modulate each other's expression by competing for miRNA binding, thereby functioning as competing endogenous RNAs (ceRNAs) [34] or acting as molecular “sponges” that sequester the miRNA [35]. Accordingly, these interactions support the construction of a predicted ceRNA network encompassing circRNAs, their related miRNAs, and downstream target mRNAs.

However, the “circRNA–miRNA sponge” hypothesis should be interpreted cautiously [36]. Most circRNAs (except a few highly abundant species such as CDR1as) are expressed at low levels, whereas miRNAs can be far more abundant [36].

Beyond miRNA sequestration, circRNAs can bind RNA-binding proteins (RBPs), influencing RNA stability, transport, and translation, and potentially modulating the function of associated proteins [37,38]. Nuclear circRNAs, can also interact with transcriptional machinery (e.g., RNA polymerase II) to regulate gene expression [39]. For example, circFOXO3 represses cell cycle progression by forming a complex with p21 and CDK2 [40]. Similarly, circYap binds TPM4 and ACTG, enhances their interaction, inhibits actin polymerization, and promotes cardiac fibrosis [41].

CircRNAs may also regulate host-gene expression by competing with linear splicing, as reported for circMBNL [42]. Finally, in mammalian muscle it has been observed that some circRNAs can be translated into proteins [43]. Several circulating microRNA (miRNAs) and lncRNAs have been associated with adverse plaque characteristics [[44], [45], [46]]. By contrast, the investigation of circRNAs as potential biomarkers of plaque vulnerability is still in its early stages. The present study explores the role of circulating circRNAs as predictors of carotid plaque vulnerability.

2. Methods

2.1. Enrollment of patients with carotid artery stenosis

Adult patients (>18 years old) who provided written informed consent were consecutively enrolled between July 18, 2019 and March 9, 2023.

Eligible patients were referred to the Vascular Surgery Unit of IRCCS Policlinico San Donato for carotid endarterectomy due to significant carotid artery stenosis. Enrollment criteria included asymptomatic patients with stenosis >80% and symptomatic patients (neurological events) with stenosis >70%, in accordance with the guidelines of the European Carotid Surgery Trial [47].

Patient data included clinical history, medications, and comorbidities. Patient characteristics are provided in Table S1.

All biological samples obtained from subjects enrolled in this study were processed and stored in the BioCor Biobank of IRCCS-Policlinico San Donato member of BBMRI-ERIC/BBMRI.it (https://www.bbmri-eric.eu/) network.

2.2. Ultrasound image acquisition and surgical procedure

Carotid stenosis was diagnosed by duplex ultrasound of the supra-aortic trunks using a MyLab Eight scanner equipped with a 7.5 MHz linear probe, model L4-15 (Esaote S.p.A.). Representative color duplex ultrasound images of stable and vulnerable carotid plaques are presented in Fig. S1.

Carotid endarterectomy was performed through a longitudinal incision of the carotid bulb extending into the internal carotid artery to expose and remove the plaque [7]. Following the removal of the adventitia, the specimen typically included the intimal layer with fibrous cap, necrotic/hemorrhagic/ulcerated material, and tunica media. The collected plaques were defined as vulnerable or not vulnerable after intraoperative macroscopic and microscopic examination of the specimens. According to the criteria described by Lovett et al., [8] and as previously reported [7], vulnerable plaques were defined by the presence of at least 1 feature among: large intraplaque hemorrhage, ulceration, large necrotic/lipidic core (∼25% of total area), ruptured or thin (<65 μm) fibrous cap, inflammatory cells infiltration, neovascularization of the plaque. All procedures adhered to the principles of the Declaration of Helsinki.

2.3. Plaque RNA isolation and RT-qPCR

Atherosclerotic plaques were transferred to the laboratory within 30 min of excision. Using a dissecting microscope (Leica™ M80, Fisher Scientific), the plaque core was separated from surrounding non-plaque tissue. Samples were divided, placed in RNAlater (Thermo Fisher Scientific) at 4 °C overnight, then removed from RNAlater and stored at −80 °C.

Total RNA was extracted using the RNeasy® Lipid Tissue Mini Kit (Qiagen), which includes on-column DNase digestion, as previously reported [7]. RNA yield and purity were assessed spectrophotometrically with a NanoPhotometer® NP80 (Implen). A total of 400 ng RNA was reverse-transcribed into cDNA using the QuantiTect Reverse Transcription Kit (Qiagen). Relative expression levels were calculated with the 2^-ΔΔCt method [48], normalized to the average Ct of Actin beta (ACTB) and Ribosomal Protein L23a (RPL23A).

2.4. PBMC sample collection, PBMC RNA isolation and RT-qPCR

Peripheral blood (4 mL) was collected immediately prior to anesthesia in K3-ethylenediaminetetraacetic acid (EDTA) tubes and stored at 4 °C for less than 4 h before processing. PBMCs were isolated by density-gradient centrifugation (1200×g, 10 min, room temperature) using Ficoll Histopaque Plus (Cytiva Sweden AB) and SepMate-15 tubes (STEMCELL Technologies), as described previously [49].

Total RNA was extracted from PBMCs with TRIzol (Life Technologies), and RNA concentration and purity were assessed spectrophotometrically. From each sample, 200 ng of RNA was reverse-transcribed into cDNA (GoScript RT Kit, Promega), followed by qPCR amplification using GoTaq Master Mix (Promega) on a CFX Opus instrument (Bio-Rad). Divergent primers targeted circRNAs, while convergent primers targeted linear host genes (Table S2). Relative expression was calculated using the 2^-ΔΔCt method [48], normalized to RPL23A.

2.5. Construction of the circRNA-miRNA-mRNA network

The circRNA-miRNA interactions were identified using the Encyclopedia of RNA Interactomes (ENCORI, https://rnasysu.com/encori/) database [50], which integrates large-scale Argonaute CLIP-Seq data.

For differentially expressed circRNAs, miRNA Recognition Elements (MREs) were interrogated. Predictions of miRNA targets were obtained from MirDIP 4.1 (https://ophid.utoronto.ca/mirDIP/) [51], which integrates predictions across multiple resources, setting the filters “high confidence” and “prediction by ≥ 10 sources” [52]. Enrichment analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was performed using ShinyGO v0.741 [53] with default parameters.

2.6. Statistical analysis

Graphs representing continuous variables are shown as median and interquartile range, while categorical variables are presented as counts and proportions and continuous variables are presented as mean ± SD or median [IQR]. Baseline characteristics by outcomes occurrence were investigated with the chi-square test or Fisher's exact test for categorical variables; continuous variables were compared by analysis of variance or by the nonparametric Mann–Whitney U test for continuous non-normally distributed data. Correlation analyses employed Spearman's or Pearson's test, as appropriate.

Multivariable logistic regression was used to identify factors that were predictors of plaque vulnerability. Age and sex were included in the multivariable model. Odds ratios (OR) with the corresponding 95% confidence interval were calculated and p values were considered statistically significant if they were less than 0.05.

The discrimination ability of circRNA scores was evaluated by ROC analyses. The statistical significance of the difference between areas under ROC curves was calculated with the method of DeLong et al. [54]. Statistical analyses were performed with SAS software, version 9.4 (SAS Institute, Inc., Cary, NC, USA) and with GraphPad Prism v.7.01 software (GraphPad Software Inc.). Graphs were designed by DataTab (https://numiqo.com) and GraphPad Prism v.7.01 software (GraphPad Software Inc.)

3. Results

3.1. circRNA deregulation in PBMCs of carotid stenosis patients

To assess the potential of circRNAs as biomarkers of carotid plaque vulnerability, we examined their expression in PBMCs, a readily accessible cell source that can, at least in part, reflect the molecular environment of surrounding tissues. Based on prior literature, four circRNAs with established relevance to endothelial function and atherosclerosis were selected for analysis [55,56].

Current circRNA naming is often unclear because back-splicing produces multiple circRNAs. Therefore, a standardized system using the prefix ‘circ’ followed by the host gene symbol and exon details was applied [57].

Specifically, we quantified the expression of these circRNAs, along with their linear host gene counterparts, in PBMCs from patients with stable (n = 102) or vulnerable (n = 85) carotid plaques. The circRNAs examined were Vir-like m6A methyltransferase associated circRNA (circVIRMA(3-4), also known as hsa_circ_0006896 or circKIAA1429) [55]; granulin circRNA (circGRN(12-13), hsa_circ_0044073) [56]; and two circular RNAs derived from the antisense non-coding RNA in the INK4 locus (circANRIL, corresponding to circular transcripts hsa_circ_0008574 or circANRIL(5-7) and hsa_circ_0004297 or circANRIL(4-14)) [58]. The chromosomal location and structure of these circRNAs are illustrated in Fig. S2-3.

For each target gene, we designed divergent primers to amplify the circular RNA and convergent primers for the linear mRNA (Table S2). The PBMC expression levels of circular RNA, the corresponding linear RNA, and the circular/linear RNA ratio are shown, allowing us to distinguish circRNA-specific regulation from overall gene expression changes.

Focusing first on circVIRMA(3-4), which is generated by back-splicing of the VIRMA gene [55] (Fig. S2A), we observed a modest but significant increase in circVIRMA(3-4) levels in PBMCs of vulnerable-plaque patients compared with stable-plaque patients (Fig. 1A). In contrast, linear VIRMA mRNA levels were not significantly different between the two groups (Fig. 1B). Consequently, the circVIRMA(3-4)/linear VIRMA mRNA ratio was higher in the vulnerable group (Fig. 1C), indicating a specific upregulation of the circular form.

Fig. 1.

circVIRMA(3-4) expression is increased in PBMCs from patients with vulnerable vs. stable carotid plaques. Total RNA was extracted from PBMCs of patients with vulnerable (n = 85) or stable (n = 102) carotid plaques. Graphs show relative expression of circVIRMA(3-4) (A), VIRMA linear mRNA (B), and circVIRMA(3-4)/VIRMA linear mRNA ratio (C) measured by RT-qPCR and expressed as log₂ fold changes. (D) The ROC curve shows the sensitivity and specificity of circVIRMA(3-4) to distinguish vulnerable from stable plaque patients. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

Similarly, circGRN(12-13), formed by circularization of two exons in the GRN gene [56] (Fig. S2B), was significantly upregulated in PBMCs from patients with vulnerable plaques (Fig. 2A), while linear GRN mRNA and the circGRN(12-13)/linear GRN mRNA ratio were not changed significantly (data not shown).

Fig. 2.

circGRN(12–13) levels are increased in PBMCs from patients with vulnerable carotid plaques. Total RNA was extracted from PBMCs of patients with vulnerable (n = 84) or stable (n = 102) carotid plaques. (A) Graph shows relative expression of circGRN(12-13) measured by RT-qPCR and expressed as log₂ fold change. (B) The ROC curve shows the sensitivity and specificity of circGRN(12-13) to distinguish vulnerable from stable plaque patients. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

ROC analysis using circVIRMA(3-4) or circGRN(12-13) expression levels showed moderate power to discriminate vulnerable from stable plaque patients (Fig. 1, Fig. 2B).

For ANRIL [58], a long noncoding RNA generated by the INK4 locus that also produces circular isoforms, we evaluated two specific circANRIL transcripts: one spanning exons 5–7 (circANRIL(5-7) also known as hsa_circ_0008574) and another spanning exons 4–14 (circANRIL(4-14)), also known as hsa_circ_0004297) (Fig. S3A and B). Neither of these circANRIL isoforms, their host gene and their circular/linear RNA ratios showed significant differential expression between vulnerable and stable groups (Figs. S4-S5).

Overall, the differential expression of these circRNAs in PBMCs was only modest. In view of these considerations, we extended our analysis to an additional circRNA, cerebellar degeneration-related protein 1 antisense (CDR1as(1), also known as hsa_circ_0001946 or CiRS7. CDR1as(1) has been more extensively studied in hypoxic cardiovascular conditions such as myocardial infarction and heart failure [[59], [60], [61], [62]], but indications of its involvement in endothelial dysfunction and atherosclerosis have been reported [62]. CDR1as(1) is generated by back-splicing of a single exon of the CDR1 gene on chromosome Xq27.1, forming a circular RNA of 1485 nucleotides (Fig. S2C). Due to limited RNA availability, CDR1as(1) expression was measured in a subset of our cohort (stable plaque n = 86; vulnerable plaque n = 65). A power calculation based on the observed variance in CDR1as(1) expression (mean log₂ fold change: stable = 0 ± 1.90, vulnerable = 1.4 ± 1.30) indicated that a sample size of 48 patients/group would provide 95% power at α = 0.01 to detect a difference. Thus, our available sample size for CDR1as(1) (86 and 65 patients per group), although reduced, was still informative. The clinical characteristics of this subset did not differ from the overall study population, and no significant clinical differences were found between the stable and vulnerable groups within the subset (Table S3).

CDR1as(1) was markedly upregulated in PBMCs from vulnerable plaque patients compared to those from stable plaque patients (Fig. 3A). In contrast, linear CDR1 RNA expression was not significantly different between the two groups (Fig. 3B). As expected, the CDR1as(1)/CDR1 mRNA ratio was greatly increased in the vulnerable group (Fig. 3C), confirming that the observed change was specific to the circular RNA. Notably, CDR1as(1) and CDR1as(1)/CDR1 RNA ratio demonstrated the strongest predictive value among the circRNAs studied, yielding the highest AUC value in ROC analysis (Fig. 3D and E). In view of the fact that CDR1as(1) derives from CDR1 gene on chromosome Xq27.1, to address the possibility of sex-related differences, we performed a sex-stratified analysis of the qPCR expression levels of circRNA CDR1as, and circVIRMA(3-4) and no statistically significant differences were observed between male and female patients (data not shown).

Fig. 3.

Increased CDR1as(1) levels in PBMCs from patients with vulnerable carotid plaques. Total RNA was extracted from PBMCs of patients with vulnerable (n = 68) or stable (n = 90) carotid plaques. Graphs show relative expression of CDR1as(1) (A), linear CDR1 mRNA (B), and CDR1as(1)/CDR1 mRNA ratio (C) measured by RT-qPCR and expressed as log₂ fold change. The ROC curve shows the sensitivity and specificity of circCDR1as(1) (D) and CDR1as(1)/CDR1 RNA ratio (E) to distinguish vulnerable from stable plaque patients. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

Likewise, no statistically significant associations were observed between circRNA expression levels and the presence of concomitant diseases (data not shown).

3.2. circRNA_SCORE and circ/lin_mRNA_SCORE elevation in PBMCs of patients with vulnerable plaques

To integrate the information from multiple circRNAs, we calculated a combined circRNA score. We defined a circRNA_SCORE by averaging the expression (log₂ fold change) of those circRNAs having a differential expression of both circRNA and circ/linear mRNA ratio values in PBMCs (specifically, circVIRMA(3-4) and CDR1as(1)). Similarly, we computed a circ/lin mRNA_SCORE by averaging the circRNA/linear mRNA expression ratios for the corresponding loci. Both circRNA_and circ/lin RNA_SCORE resulted significantly elevated in PBMCs from patients with vulnerable plaques compared to those with stable plaques (Fig. 4A and B). No statistically significant associations were observed between circRNA_SCORE or circ/lin_SCORE and the presence of concomitant diseases or between males and females (data not shown).

Fig. 4.

circRNA_SCORE and circRNA/lin mRNA ratio_SCORE levels in PBMCs from patients with carotid artery stenosis. The expression values of CDR1as(1) and circVIRMA(3-4) or of CDR1as(1)/CDR1 mRNA ratio and circVIRMA(3-4)/VIRMA linear mRNA ratio were averaged to calculate the circRNA_SCORE (A) (vulnerable n = 60; stable n = 85) or the circRNA/lin mRNA ratio_SCORE (B) (vulnerable n = 59; stable n = 78). Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05). The ROC curves show the sensitivity and specificity of circRNA_SCORE (C) and circRNA/lin mRNA ratio_SCORE (D) to distinguish vulnerable from stable plaque patients.

In predictive performance, the circRNA_SCORE and circ/lin_SCORE demonstrated good sensitivity and specificity for distinguishing vulnerable from stable plaques (Fig. 4C); however, they did not perform significantly better than CDR1as(1) or CDR1as(1)/CDR1 mRNA ratio alone, even after adjusting for age and sex, as shown in the multivariate analysis with ROC curves comparison analysis (Table S4).

The diagnostic performance of individual circRNA markers and composite scores was evaluated using maximum sensitivity and specificity at the optimal cut-off values (Table S5). Among the tested markers, CDR1as(1)/CDR1 mRNA ratio showed the best overall balance (specificity 80%, sensitivity 69%, cut-off 0.4). CDR1as(1) and circRNA_SCORE had the highest sensitivity (81%) but only moderate specificity (61% and 59%), making them more suitable for screening. In contrast, circ/linear mRNA ratio_SCORE and circVIRMA(3–4) had higher specificity (76% and 66%) but lower sensitivity (57% and 51%), suggesting more value for confirming rather than detecting cases.

3.3. Increase of the CDR1as(1)/lin CDR1 mRNA ratio levels in vulnerable plaque tissue

We assessed the expression of the studied circRNAs in carotid plaque tissue specimens (plaque core samples) by qPCR. Unfortunately, due to technical limitations, the quality and quantity of RNA extracted from carotid plaque tissue were not consistently sufficient to perform RT-qPCR experiments. As a result, the number of patients included in the cytokine/PPARgamma and CDR1as measurements from plaque tissues differs from the number used for PBMC measurements. Among the five circRNAs tested, only CDR1as(1) was reliably detectable in the plaque tissue. When comparing plaque tissues, CDR1as(1) levels showed a slight and non-significant decrease in vulnerable plaques relative to stable plaques (Fig. S6A), whereas linear CDR1 mRNA was more substantially and significantly reduced in vulnerable plaques (Fig. S6B). As a result, the CDR1as(1)/CDR1 ratio was significantly higher in vulnerable plaque tissue (Fig. S6C), suggesting a circRNA-specific regulation within the plaques.

3.4. Inflammatory marker characterization of the carotid plaques

Vulnerable atherosclerotic plaques are characterized by elevated proinflammatory activity of T cells and macrophages, resulting in increased cytokine expression [3]. Consistent with this, we examined the messenger RNA (mRNA) levels of key cytokines in carotid plaque tissues. Total RNA was extracted from carotid endarterectomy specimens (42 vulnerable plaques and 61 stable plaques) and analyzed by RT-qPCR for a panel of cytokines/chemokines. Specifically, we quantified mRNA levels of tumor necrosis factor alpha (TNF-α), C-C motif chemokine ligand 5 (CCL5, also known as RANTES), interleukin-1β (IL-1β), interleukin-6 (IL-6), and interleukin-10 (IL-10). As shown in Fig. 5A–C, IL-1β, IL-6, and IL-10 mRNA levels were significantly higher in vulnerable plaques compared with stable plaques. We also measured the expression of peroxisome proliferator–activated receptor gamma (PPARγ), a nuclear receptor implicated in atherosclerosis [[63], [64], [65]], and found that it was elevated in vulnerable plaques (Fig. 5D). In contrast, TNF-α and CCL5 levels did not differ significantly between the groups (data not shown). Notably, IL-10 and PPARγ expression levels showed a significant positive correlation (Fig. S7A).

Fig. 5.

IL-1β, IL-6, IL-10 and PPARγ expression levels are elevated in vulnerable carotid plaques and in-plaque_SCORE is correlated with circRNA_SCORE. Total RNA was extracted from plaque tissue of patients (vulnerable n = 41, stable n = 59) with carotid artery stenosis. Dot-plots show relative gene expression of IL-1β (A), IL-6 (B), IL-10 (C) and PPARγ (D) measured by qPCR and expressed as log₂ fold change. Data are represented as dot-plots, box-and-whiskers and the probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05). (E) circRNA_SCORE is correlated with in-plaque_SCORE in patients with vulnerable carotid plaques. The fold change of circRNA_SCORE and in-plaque_SCORE were analyzed by Spearman's correlation test. Spearman's correlation coefficient and statistical significance are indicated.

To derive an overall measure of inflammatory activity within plaques, we averaged the log₂ fold change values of the three differentially expressed cytokines (IL-1β, IL-6, IL-10) and PPARγ. This composite metric, termed the in-plaque_SCORE, was significantly higher in vulnerable plaques than in stable plaques (Fig. S7B). Interestingly, in a subset, due to technical limitations, of patients with vulnerable plaques, the circRNA_SCORE was positively correlated with the in-plaque_SCORE derived from cytokine and PPARγ expression (Fig. 5E), linking systemic circRNA changes to plaque inflammatory status.

3.5. Vulnerable plaque ceRNA regulatory network and predicted functional pathways

To explore the potential regulatory interactions underlying plaque vulnerability, we constructed a putative competing endogenous RNA (ceRNA) network involving the circRNAs and their associated miRNAs and target mRNAs. MiRNAs repress gene expression by binding to specific microRNA response elements (MREs) on target transcripts [33]. If circRNAs and mRNAs share common MREs for a given miRNA, they can influence each other's expression by competing for that miRNA, effectively acting as ceRNAs [34] or molecular “sponges” that sequester the miRNA [35].

Using the ENCORI database, we identified a total of 150 circRNA–miRNA interactions for the three circRNAs that we found deregulated in PBMCs. We then performed target prediction for these miRNAs using mirDIP, which yielded 18,421 putative target mRNAs (predicted by ≥ 10 different algorithms). These results are summarized in Table S6, with the full list of predicted targets available in Supplementary file S1. The unique predicted mRNAs were subjected to KEGG pathway enrichment analysis to identify biological pathways potentially affected in patients with vulnerable plaques. Among the top 20 significantly enriched KEGG pathways were the RAS signaling, RAP1 signaling, and Focal adhesion pathways (Supplementary file S2 and Fig. S8). These pathways are all related to cell signaling and adhesion and have known roles in vascular biology and atherosclerosis.

4. Discussion

Molecular changes within the carotid arterial wall are strongly linked to plaque vulnerability. Inflammatory mediators—including cytokines, chemokines, and adhesion molecules—play a key role by promoting endothelial dysfunction, leukocyte recruitment, and foam cell formation by smooth muscle cell–derived macrophage-like cells, as well as by endothelial cells [3].

In this study, we demonstrated that vulnerable plaques exhibit higher expression of pro-inflammatory cytokine genes such as IL-1β and IL-6, as well as anti-inflammatory mediators including IL-10 and PPARγ [38,51]. The positive correlation between IL-10 and PPARγ is consistent with evidence that PPARγ exerts anti-inflammatory activity partly via induction of IL-10 [66]. By integrating expression levels of IL-1β, IL-6, IL-10, and PPARγ into a composite metric, the in-plaque_SCORE, we showed that both pro- and anti-inflammatory responses are enhanced in vulnerable plaques, indicating the coexistence of pathological drivers of instability and adaptive counter-regulatory mechanisms.

The study aimed to identify circulating biomarkers that can reliably predict plaque vulnerability, as current biomarkers lack sufficient sensitivity, specificity, and reproducibility for clinical application [22,23]. Advances in RNA sequencing have expanded knowledge of noncoding RNAs in CVDs [25,67], with circular RNAs (circRNAs) emerging as promising but still underexplored regulatory molecules and potential biomarkers in CVDs [25,[27], [28], [29], [30]].

We initially investigated circRNAs previously associated with endothelial biology and atherosclerosis [55,56]. CircVIRMA(3-4) expression was modestly but significantly increased in PBMCs from patients with vulnerable plaques, while linear VIRMA was unchanged. CircVIRMA(3-4) has been reported to be enriched in serum-derived exosomes of patients with vulnerable carotid atherosclerosis, correlating with triglycerides, low-density lipoprotein cholesterol (LDL-C), and C-reactive protein (CRP) [55]. Exosomal circVIRMA also promoted endothelial proliferation and migration [55], at least in part, via downregulation of hsa-miR-1264 and consequent upregulation of its target DNA (cytosine-5)-methyltransferases (DNMT1) [68]. Since DNMT1-driven DNA methylation changes are implicated in flow-induced endothelial dysfunction [69,70], circVIRMA may contribute to vascular remodeling in atherosclerosis.

CircGRN(12-13) was likewise elevated in PBMCs from vulnerable patients. Previous studies found circGRN(12-13) increased in serum of patients with coronary atherosclerosis and demonstrated that it promotes proliferation and inflammation in vascular smooth muscle and endothelial cells via the hsa-miR-107/JAK/STAT axis [56]. Together, these findings suggest that circVIRMA(3-4) and circGRN(12-13) are functionally involved in vascular inflammation and remodeling.

We also examined circANRIL isoforms from the 9p21 locus, a region strongly linked to atherosclerosis [58,71,72]. Linear ANRIL is associated with increased disease risk, whereas circANRIL isoforms exert protective effects [58,71,72]. In our cohort, neither circANRIL isoforms (circANRIL(5-7), circANRIL(4-14)) nor linear ANRIL showed differential expression between groups, indicating that PBMC levels of circANRIL are not major discriminators of carotid plaque vulnerability in this setting.

Given the modest discriminatory power of circVIRMA(3-4) and circGRN(12-13), we expanded our analysis to CDR1as(1) (CiRS-7), a highly conserved circRNA that functions as a hsa-miR-7 sponge [[73], [74], [75]]. CDR1as(1) has been primarily studied in ischemic cardiovascular conditions, including myocardial infarction and heart failure [[59], [60], [61], [62]]. Elevated CDR1as(1) levels have been observed in whole blood of acute myocardial infarction patients [59], where combined measurement with the lncRNA ZFAS1 improved predictive performance. Increased plasma CDR1as(1) has also been reported in heart failure, correlating with BNP levels [62,76].

Our findings demonstrate that CDR1as(1) is significantly upregulated in PBMCs of patients with vulnerable carotid plaques. Both CDR1as(1) expression and the CDR1as(1)/CDR1 ratio showed strong predictive capacity, outperforming other circRNAs tested. These results align with reports that circulating CDR1as(1) is elevated in coronary artery disease and serves as an independent indicator of risk [77]. Together, these observations highlight CDR1as(1) as a promising biomarker linking systemic inflammation, atherosclerotic burden, and ischemic risk.

To assess circRNA data, a circRNA_SCORE and a circRNA/lin RNA ratio_SCORE were calculated from the average log₂ fold change of differentially expressed circRNAs and circRNA/linear mRNA ratios. These scores were higher in PBMCs from patients with vulnerable plaques and correlated with the in-plaque_SCORE, indicating systemic reflection of local inflammation. While predictive in analyses, the circRNA_ and the circRNA/lin RNA ratio_SCORE did not outperform CDR1as(1), highlighting CDR1as(1) or CDR1as(1)/CDR1 mRNA ratio as strong standalone biomarkers. Moreover, the diagnostic performance of individual circRNA markers and composite scores was evaluated and the best balanced marker overall resulted the CDR1as(1)/CDR1 mRNA ratio.

To investigate mechanisms behind circRNA deregulation, a circRNA–miRNA–mRNA interaction network was built. Enrichment analysis revealed key pathways—Rat sarcoma virus protein (RAS) signaling, Ras-related protein 1 (RAP1) signaling, and focal adhesion—as top KEGG terms. These pathways are fundamental to vascular biology: RAS signaling influences endothelial growth, adhesion, migration, and vascular smooth muscle cell inflammation and contributes to VSMC senescence and inflammation [78,79]; RAP1 signaling regulates nitric oxide release and endothelial balance, especially under shear stress [80]; focal adhesion controls integrin-mediated cell–matrix interactions crucial in vascular remodeling, restenosis, atherosclerosis, heart failure, and thrombosis [81,82]. These three enriched pathways are interconnected with each other, and all are related to atherosclerosis mechanisms. In view of these data, the network activation triggered by circCDR1as(1), circVIRMA(3-4) and circGRN(12-13) may contribute to atherogenesis. It is important to note that circRNAs can regulate gene expression through multiple mechanisms beyond acting as ceRNAs. circRNAs can bind RNA-binding proteins (RBPs), influencing RNA stability, transport, and translation, and potentially modulating the function of associated proteins [37,38]. CircRNAs can also regulate host-gene expression by competing with linear splicing [42], and finally, some circRNAs can be translated [43].

Our findings should be interpreted considering some limitations. First, validation in larger and independent cohorts is required to confirm the viability of these circRNAs as biomarkers. Second, PBMC-derived RNA profiles may vary according to patient-specific immune and inflammatory status, potentially influencing the observed transcript levels [83]. Third, the ceRNA network and KEGG pathway analyses are based on in silico predictions and should be interpreted as hypothesis-generating. Accordingly, these findings suggest potential regulatory interactions and pathway involvement, but do not demonstrate causality. Therefore, experimental validation (e.g., functional assays and targeted mechanistic studies) will be required to confirm the predicted circRNA–miRNA–mRNA interactions and the biological relevance of the enriched pathways. Fourth, the pro-/anti-inflammatory markers were assessed at the mRNA level rather than at the protein level. Since gene and protein expression are not always quantitatively aligned, this should be considered when interpreting these results. Fifth, the potential coexistence of vulnerable and stable plaques within the same individual represents an inherent limitation, shared by most biomarker studies in atherosclerosis. Circulating biomarkers capture the patient's global inflammatory and atherosclerotic burden rather than the characteristics of a specific lesion. As a result, classifying patients based on a single excised plaque may lead to some degree of misclassification, especially when plaque heterogeneity exists across different vascular territories.

5. Conclusion

In conclusion, we identified three circRNAs—CDR1as(1), circVIRMA(3-4), and circGRN(12-13)—as potential circulating biomarkers of carotid plaque vulnerability. Among these, CDR1as(1) exhibited the strongest predictive capacity, both alone and relative to composite scores. Importantly, circRNA expression correlated with intraplaque inflammatory signatures, supporting their role as systemic indicators of local pathology. These findings provide new insights into the molecular determinants of plaque instability and highlight circRNAs as promising candidates for noninvasive risk stratification in carotid artery disease, including asymptomatic patients.

CRediT authorship contribution statement

Simona Greco: Writing – review & editing, Writing – original draft, Formal analysis, Conceptualization. Federica Perego: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Conceptualization. Santiago Nicolas Piella: Investigation, Formal analysis. Federico Favretto: Investigation, Formal analysis. Spyros Tastsoglou: Methodology, Data curation. Valentina Milani: Methodology, Data curation. Sara Boveri: Methodology, Data curation. Laura Valentina Renna: Writing – review & editing, Methodology. Rosanna Cardani: Writing – review & editing, Methodology. Federico Ambrogi: Writing – review & editing, Methodology, Data curation. Giovanni Nano: Writing – review & editing, Methodology, Data curation. Nicoletta Basilico: Writing – review & editing, Writing – original draft, Data curation, Conceptualization. Daniela Mazzaccaro: Writing – review & editing, Writing – original draft, Supervision, Methodology, Funding acquisition, Data curation, Conceptualization. Fabio Martelli: Writing – review & editing, Writing – original draft, Supervision, Funding acquisition, Conceptualization.

Ethics approval

The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki and was approved by the Ethics Committee of San Raffaele Hospital (Italy) on June 20, 2019 (110/int/2019; ClinicalTrials.gov: NCT05566080) and on July 18, 2019 (131/int/2019; ClinicalTrials.gov: NCT06217471).

Data availability statement

The raw data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declaration of generative AI and AI-assisted technologies in the manuscript preparation process

During the preparation of this work the authors used ChatGPT (OpenAI, San Francisco, CA) in order to improve the grammar and clarity of the manuscript. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.

Funding

F.M. is partially supported by the Italian Ministry of Health (POS-T4 CAL.HUB.RIA T4-AN-09), and by the European Union (Next Generation EU-NRRP M6C2 Inv. 2.1 PNRR-MAD 2022-12375790 and PNRR-MCNT2-2023-12377983, and Romania’s PNRR-III-C9-2022-I8, CF 186/24.11.2022, contr. 760062/23.05.2023). F.M. S.G. are also partially supported by Ricerca Corrente funding from Italian Ministry of Health to IRCCS Policlinico San Donato (#1.07.128, #1.07.129 and #1.05.111). LVR was supported by Ricerca Corrente 1.15.130, and RC by Ricerca corrente 6.01.101.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the supplementary data to this article:

figs1.

Representative color duplex ultrasound images of carotid plaques. A) Significant carotid stenosis indicated by white arrow due to a stable carotid plaque. Note the iso/iperechogenic aspect of the atheroma with regular surface. B) a vulnerable carotid plaque, described by the hypo/anechoic aspect with irregular surface of the atheroma. Ultrasound images were obtained by duplex ultrasound of the supra-aortic trunks using a MyLab Eight scanner equipped with a 7.5 MHz linear probe, model L4-15 (Esaote S.p.A.).

figs2.

CircRNAs genomic annotation and circularization scheme. (A) circVIRMA(3-4), (B) circGRN(12-13), (C) CDR1as(1). UCSC Hg19 Genome Browser annotation of circRNA isoforms generated by VIRMA, GRN, and CDR1 genes and present in CircBase database are reported. The investigated circular isoforms are indicated by red squares.

figs3.

CircANRILs genomic annotation and circularization scheme. The UCSC Hg19 Genome Browser annotation of circular RNAs hosted by ANRIL gene as represented in CircBase database (black) are reported. The two investigated circANRIL isoforms are indicated by red and blu squares.

figs4.

circANRIL(5-7) and ANRIL expression levels in PBMC of carotid artery stenosis patients. Total RNA was extracted from PBMCs from patients with vulnerable (n = 85) or stable (n = 102) carotid plaques. Graphs show the relative expression of circANRIL(5-7) (A), linear ANRIL RNA (B) and circANRIL(5-7)/ANRILlinear RNA ratio in (C) measured by RT-qPCR and expressed as log₂ fold change. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

figs5.

circANRIL(4-14) and ANRIL expression levels in PBMC of carotid artery stenosis patients. Total RNA was extracted from PBMCs from patients with vulnerable (n = 85) or stable (n = 102) carotid plaques. Graphs show the relative expression of circANRIL(4-14) (A) linear ANRIL RNA (B) and circANRIL(4-14)/ANRIL linear RNA ratio in (C), measured by RT-qPCR and expressed as log₂ fold change. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

figs6.

CDR1as(1)/CDR1 ratio is increased in vulnerable plaque tissue. Total RNA was extracted from plaque tissue (vulnerable n = 52, stable n = 45) of patients with carotid artery stenosis. Graphs show relative expression of CDR1as(1)(A), CDR1 (B) and CDR1as(1)/CDR1 ratio (C) measured by RT-qPCR and expressed as log₂ fold change. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

figs7.

PPARg/IL-10 correlation and in-plaque_SCORE increase in vulnerable carotid plaques. Total RNA was extracted from stable (n = 59) and vulnerable (n = 41) plaques of patients with carotid artery stenosis. (A) In-plaque expression levels of PPARg and IL-10 are correlated in patients with carotid artery stenosis. The fold change expression of PPARg and IL-10 was analyzed by Spearman’s correlation test. Spearman’s correlation coefficient and statistical significance are indicated. Blue and red dots indicate patients with stable or vulnerable plaques, respectively. (B) IL-1β, IL-6, IL-10 and PPARg mRNA log₂ fold changes in vulnerable or stable carotid plaques were averaged and expressed as in-plaque_SCORE. Graphs show the in-plaque_SCORE levels increased in vulnerable carotid plaques. Data are represented as dot-plots, box-and-whiskers and probability distribution violin plots. Medians with interquartile ranges are also shown. Statistical significance was assessed using the Mann–Whitney U test (statistical significance threshold, p < 0.05).

figs8.

Vulnerable plaque ceRNA regulatory network and predicted functional pathways. Pathways are represented by lollipop graph ordered by FDR significance. The size of the lollipop indicates the number of genes, and colors indicate the statistical significance (red for lower FDR, blue for higher FDR).